Subscale Tiltrotor eVTOL Aircraft Dynamic Modeling and Flight Control Software Development

Subscale Tiltrotor eVTOL Aircraft Dynamic Modeling and Flight Control

Software Development

Benjamin M. Simmons Kasey A. Ackerman Research Aerospace Engineer Research Aerospace Engineer Flight Dynamics Branch Dynamic Systems and Control Branch Garrett D. Asper Matthew N. Gray Pathways Intern Electronics Engineer Flight Dynamics Branch Aeronautics Systems Engineering Branch Steven M. Snyder Rachel M. Axten Research Aerospace Engineer Pathways Intern Dynamic Systems and Control Branch Dynamic Systems and Control Branch Steven C. Geuther Ryan Chan Aerospace Engineer Electronics Engineer Aeronautics Systems Engineering Branch Aeronautics Systems Engineering Branch NASA Langley Research Center Hampton, VA, USA ABSTRACT This paper describes the dynamic modeling and flight control software development efforts for a subscale tiltrotor electric vertical takeoff and landing (eVTOL) aircraft built at NASA Langley Research Center. The vehicle, referred to as the Research Aircraft for eVTOL Enabling techNologies (RAVEN) Subscale Wind-Tunnel and Flight Test (SWFT) model, serves as a flight dynamics and controls research testbed to foster advances in eVTOL aircraft technology.

After fabricating the vehicle, wind-tunnel testing was conducted to identify a high-fidelity aero-propulsive model for use in a flight dynamics simulation enabling flight control system development. The RAVEN-SWFT aircraft subsequently underwent flight-test risk reduction steps and then free flight testing employing custom research flight control software. The flight control software, which can be efficiently updated and tested on the vehicle, includes a robust model-based control algorithm and an extensive programmed test input injection capability. The progress of RAVEN-SWFT research activities will be summarized alongside the associated modeling and flight control software development aspects, including the flight dynamics simulation, flight control system architecture, vehicle integration, and testing approaches.

Vehicles designed for UAM missions require precise hover INTRODUCTION and efficient cruise capabilities, as well as the ability to safely transition between flight regimes; electric VTOL aircraft are a Many complex distributed hybrid and electric propulsion air- promising candidate to fulfill future UAM operations. In gen- craft concepts are being investigated to enable future Ad- eral, eVTOL aircraft are a combination of traditional fixed- vanced Air Mobility (AAM) transportation missions (Refs. 1– wing and rotary-wing aircraft leveraging certain attributes 6). There are numerous vertical takeoff and landing (VTOL), from each type of vehicle. Fixed-wing aircraft provide longer short takeoff and landing (STOL), and conventional takeoff endurance, better efficiency, and the ability to operate at high and landing (CTOL) configurations with vast Urban Air Mo- speeds, whereas rotary-wing aircraft offer the ability to take- bility (UAM) and Regional Air Mobility (RAM) applications.

off and land vertically, hover, and precisely maneuver in con- Ubiquitous characteristics of these novel aircraft include the fined areas. New distributed propulsion technology that is use of many distributed propulsors and control surfaces, as used in many eVTOL vehicles has further expanded the tra- well as significant aero-propulsive coupling.

ditional aircraft design space and has yielded a multitude of unique vehicle designs. As of March 2025, there are 1,100 Presented at the Vertical Flight Society’s 81st Annual Forum & Technology Display, Virginia Beach, VA, USA, May 20–22, 2025.

known eVTOL aircraft concepts (Ref. 7).

This material is declared a work of the U.S. Government and is not subject to copyright protection in the United States. Although widespread use of eVTOL aircraft has great po- tential, many open research areas need to be addressed TOL configuration with six variable-pitch proprotors. The prior to introduction into a UAM transportation environment. front four proprotors tilt forward and are operational through- One essential research area is accurate eVTOL vehicle aero- out the entire flight envelope. The rear two proprotors do not propulsive modeling enabling flight dynamics simulation de- tilt and serve as lifting proprotors in hover and transition. The velopment. Another critical eVTOL aircraft research area, aircraft control surfaces include six flaperons, a stabilator, and which substantially benefits from an accurate flight dynamics a rudder. In total, the vehicle has 24 independent control ef- simulation, is flight control system development. Together, fectors: the delivery of a high-fidelity flight dynamics model and ef- • Six proprotor rotational speeds ( 𝑛 , 𝑛 , ..., 𝑛 ) 1 2 6 fective flight control system enable both accurate flight simu- • Six proprotor collective pitch angles ( 𝛿 , 𝛿 ,..., 𝛿 ) 𝑐 𝑐 𝑐 1 2 6 lations and successful flight testing, which facilitates research • Four nacelle tilt angles ( 𝛿 , 𝛿 , 𝛿 , 𝛿 ) 𝑡 𝑡 𝑡 𝑡 1 2 3 4 in many other areas including airworthiness certification, air • Six flaperon deflection angles ( 𝛿 , 𝛿 ,..., 𝛿 ) 𝑓 𝑓 𝑓 1 2 6 traffic management, pilot-operator interface, handling quali- • One stabilator deflection angle ( 𝛿 ) 𝑠 ties, simplified vehicle operations, contingency management, • One rudder deflection angle ( 𝛿 ) 𝑟 and autonomous systems. Efficient, accurate aero-propulsive Figure 1 shows a schematic of the RAVEN aircraft with an- model development and robust, well-performing flight con- notations showing the vehicle propulsor and control surface trol system design, however, is challenged by several eV- definitions.

TOL vehicle attributes, including: many control surfaces and propulsors, propulsion-airframe interactions, vehicle instabil- 𝜹 𝒇 𝜹 𝟓 𝒕 𝟒 𝒏 , 𝜹 ity, and rapidly changing aerodynamics through large opera- 𝟔 𝒄 𝟔 𝜹 𝒓 𝜹 𝒇 𝟔 tional flight envelopes. Previous related research has investi- gated methods for rapid, high-fidelity eVTOL aircraft aero- 𝜹 𝜹 𝒇 𝒔 𝟑 propulsive modeling across their wide flight envelopes us- 𝜹 𝒇 𝟒 𝒏 , 𝜹 𝟒 𝒄 𝟒 ing computational predictions (Refs. 8, 9), wind-tunnel test- 𝜹 ing (Refs. 10–16), and flight-test approaches (Refs. 17,18). In 𝒕 𝟑 parallel, full-envelope flight control system development ap- 𝒏 , 𝜹 𝟓 𝒄 𝜹 𝟓 𝒇 𝟐 proaches for complex transitioning eVTOL aircraft have been 𝜹 𝒇 𝟏 explored (e.g., Refs. 19–24).

𝒏 , 𝜹 𝟑 𝒄 𝟑 𝜹 𝒕 𝟐 The present work describes the modeling and flight control 𝜹 𝒕 𝟏 software development methods applied for a subscale tiltro- 𝒏 , 𝜹 𝟐 𝒄 𝟐 tor eVTOL aircraft to realize the project’s research objectives.

𝒏 , 𝜹 𝟏 𝒄 𝟏 Furthermore, the control algorithm design approach and test- ing efforts are summarized. The paper strives to present the Figure 1: RAVEN control effector definitions.

software development and testing techniques with sufficient detail to allow for the methods to be used to help collectively The RAVEN-SWFT, pictured in the NASA LaRC 12-Foot advance the state-of-the-art of eVTOL aircraft technology.

Low-Speed Tunnel (LST) (Ref. 27) in Fig. 2, was designed The paper is organized as follows. First, an overview is pro- as a flight dynamics and controls research testbed to advance vided of the experimental eVTOL aircraft. Next, the wind- eVTOL aircraft technology and help bring similar full-scale tunnel testing and aero-propulsive modeling efforts are sum- vehicles into mainstream operation. As its name suggests, the marized, followed by a description of the flight dynamics vehicle is designed for use in both wind-tunnel and flight-test simulation. Subsequent sections detail the flight control sys- experiments. The aircraft has a flight weight of 38 pounds, a tem design, programmed test input injection capabilities, and wingspan of 5.7 ft (excluding the outboard tilt nacelles), and flight control system integration methods. The paper closes a proprotor diameter of 19.5 inches. The current RAVEN- with a discussion of software assurance and airworthiness SWFT moments of inertia and product of inertia estimates, ex- considerations, the flight-test approach being applied for the perimentally determined using both torsional and compound vehicle, and concluding remarks.

pendulum methods, are provided in Ref. 25.

The RAVEN-SWFT was developed as NASA LaRC’s latest AIRCRAFT and most capable subscale eVTOL research aircraft. Two pre- To help study the aforementioned eVTOL aircraft research decessor subscale eVTOL aircraft were the GL-10 (Ref. 28) areas, NASA Langley Research Center (LaRC) has built and LA-8 (Refs. 29, 30). The GL-10 was a subscale, tiltwing, the Research Aircraft for eVTOL Enabling techNologies tilt-tail VTOL aircraft. Several scaled variants of the GL-10 (RAVEN) Subscale Wind-Tunnel and Flight Test (SWFT) vehicle were developed and tested, which enabled research model (Ref. 25). The RAVEN-SWFT is a 28.6% scale in wind-tunnel testing (Refs. 10, 11), flight controls (Ref. 31), version of the RAVEN 1000-lb class eVTOL aircraft con- and flight testing (Refs. 32,33). The LA-8 was a subscale, tan- cept (Ref. 26), which was designed in a collaborative effort dem tiltwing VTOL aircraft built as a wind-tunnel and flight between NASA LaRC and the Georgia Institute of Technol- testbed for eVTOL aircraft technology. The LA-8 project ogy (Georgia Tech). The RAVEN aircraft is a tiltrotor eV- enabled research in rapid vehicle development (Refs. 29, vanced modeling and flight controls research. Accordingly, a large emphasis was placed on being able to rapidly design, integrate, test, and refine custom flight control algorithms.

A significant emphasis of the RAVEN project has been on public dissemination of methods and data to foster advance- ment of the eVTOL aircraft industry. Reference 25 provides a summary of past, current, and future research efforts be- ing pursued by the RAVEN-SWFT project, which includes isolated proprotor characterization (Ref. 42), static powered- airframe wind-tunnel testing and aero-propulsive model de- velopment (Ref. 15), computational aerodynamics predic- tions (Ref. 43), and aerodynamic damping estimation from (a) Front view (cruise configuration) free motion wind-tunnel testing (Ref. 16). The RAVEN- SWFT wind-tunnel testing efforts are summarized in the next section to provide context and explain their purpose for the present application. The current paper focuses on the modeling and flight control software development enabling flight-test research conducted using the RAVEN-SWFT ve- hicle. This substantial software development effort has only been mentioned glancingly in previous publications and, thus, forms the primary emphasis and new contributions of this pa- per.

WIND-TUNNEL TESTING AND AERO-PROPULSIVE MODELING (b) Overhead view (hover configuration) Wind-tunnel tests performed in the NASA LaRC 12-Foot LST for RAVEN-SWFT enabled identification of an accurate aero- propulsive model used for flight dynamics simulation devel- opment and model-based flight control system design. The static wind-tunnel test campaign included characterizing the isolated proprotor (Ref. 42), the isolated airframe (without proprotors), and powered airframe (with proprotors operat- ing) (Ref. 15). Design of experiments (DOE) and response surface methodology (RSM) test techniques (Refs. 44, 45) were applied throughout static wind-tunnel testing (Ref. 15).

In contrast to traditional one-factor-at-a-time testing, exper- iments planned using DOE/RSM theory efficiently scale for a large number of test factors, which is essential for charac- (c) Side view (mid-transition configuration) terizing the complex nonlinear aerodynamics and interactions present with eVTOL aircraft. DOE/RSM techniques increase Figure 2: RAVEN-SWFT mounted in the NASA LaRC 12- the productivity of data collection by simultaneously varying Foot Low-Speed Tunnel. (Credit: NASA) all test factors in a way that allows efficient determination of the individual contribution of each individual factor, as well 30), computational aerodynamic predictions (Refs. 34–36), as interaction effects among test factors, while fundamentally wind-tunnel testing (Refs. 12, 37), high incidence angle pro- providing a statistically-rigorous experiment design approach.

peller aerodynamics (Refs. 38–40), aero-propulsive model- RAVEN-SWFT is the first eVTOL vehicle tested at NASA ing (Refs. 13, 14, 41), flight controls (Ref. 19), and flight-test LaRC to primarily apply DOE/RSM techniques throughout strategies (Ref. 17). The GL-10 and LA-8 both had successful its static wind-tunnel testing.

wind-tunnel and flight-test campaigns that yielded numerous eVTOL aircraft research advances; however, due to budgetary The RAVEN-SWFT static wind-tunnel testing yielded the constraints and programmatic adjustments, neither project ful- data needed to identify a high-fidelity static transition model; filled the research goal of testing custom flight control soft- however, aerodynamic damping effects are not characterized ware throughout the full flight envelope, leaving an important in static testing and are required to accurately predict air- open flight-test research area unexplored. From its initial con- craft motion in dynamic maneuvering with nonzero angular ceptualization, the RAVEN-SWFT aircraft was designated to velocity. Forced oscillation testing (Refs. 46, 47) is a tra- continue eVTOL aircraft research pursuits started on the GL- ditional dynamic wind-tunnel test technique used to charac- 10 and LA-8, with a key focus on creating a testbed for ad- terize aerodynamic damping by forcing the dynamic motion of a wind-tunnel model; however, due to the significant air- settings that are able to be tested in the 12-Foot LST. Also, craft complexity and large number of operating conditions, accurate extrapolation of an integrated aero-propulsive vehi- applying this approach to eVTOL vehicles would be very cle model outside of the test region is difficult due to the time consuming. Instead, to increase the efficiency of dy- significant propulsion, airframe, and interactional contribu- namic wind-tunnel testing, three degree-of-freedom (3DOF) tions to the vehicle aerodynamics. To provide a prelimi- free motion wind-tunnel testing was used to characterize the nary estimate of the vehicle aero-propulsive forces and mo- RAVEN-SWFT aero-propulsive damping effects. Figure 3 ments at high speeds (i.e., at dynamic pressure values above shows the RAVEN-SWFT vehicle mounted on a 3DOF ap- 5 lbf/ft ), a low-fidelity model was created by superimposing paratus that was recently designed and built at NASA LaRC. proprotor and isolated airframe models identified from wind- In essence, the RAVEN-SWFT 3DOF wind-tunnel testing tunnel testing, which have better independent extrapolation for aero-propulsive damping characterization was executed by abilities due to their isolated nondimensionalization. This ap- having the flight control system track commanded attitude an- proach ignores interactional aerodynamics, which have a sub- gles and transition the aircraft based on tunnel dynamic pres- stantial effect; therefore, future computational and/or flight- sure, while also injecting multisine programmed test input ex- test model identification is required to produce an accurate citations to enable collection of informative data for model aero-propulsive model valid in the high-speed portion of the identification (Ref. 16). As will be discussed later in the paper, RAVEN-SWFT flight envelope.

the 3DOF testing also allowed for flight control algorithms to be rapidly tested and refined in a low-risk, flight-like environ- FLIGHT DYNAMICS SIMULATION ment.

A flight dynamics simulation was developed for model-based flight control system design and testing using the RAVEN- SWFT aero-propulsive models. The flight dynamics simula- ® tion was developed in Simulink and is a modified version of NASA’s open source “Flight Dynamics Simulation of a Generic Transport Model” (Ref. 48), where pertinent changes were made to simulate flight for the RAVEN-SWFT aircraft.

® The top level of the RAVEN-SWFT simulation Simulink di- agram is shown in Fig. 4, which is composed of the simulation inputs, flight control system, vehicle model, and simulation outputs. The inputs, outputs, and vehicle model are described in the following subsections. The flight control system (FCS) is described in the next section.

Figure 3: RAVEN-SWFT mounted on a 3DOF free mo- tion wind-tunnel apparatus in the NASA LaRC 12-Foot LST.

(Credit: NASA) After completing wind-tunnel testing, RAVEN-SWFT aero- propulsive models were created from data acquired at sev- eral dynamic pressure settings in the transition flight enve- lope up to 5 lbf/ft (the practical dynamic pressure limit for testing eVTOL vehicles in the 12-Foot LST). The form of Figure 4: Top-level view of the RAVEN-SWFT simulation the model is a set of polynomial response surface equations ® Simulink diagram.

(RSEs). The modeled responses were the dimensional body- axis aero-propulsive forces and moments predicted as a func- tion of the body-axis translational velocity, body-axis angu- Simulation Inputs lar velocity, proprotor rotational speeds, and control effector deflection angles (Refs. 15, 16). To aid in control design, The Simulation Inputs subsystem includes simulated radio the RSEs were symmetrized by constraining the identifica- control (RC) transmitter inputs, simulated ground control sta- tion procedures to force control effectors mirrored over the tion (GCS) commands, toggleable automatic arming logic for 𝑥 – 𝑧 plane to have equal effectiveness and restrict side force, batch simulations, and a simulation elapsed time counter. The rolling moment, and yawing moment to be zero at zero lateral velocity (sideslip) conditions.

The use of trademarks or names of manufacturers in this report is for accurate reporting and does not constitute an official endorsement, either expressed or Notably, the high-speed transition and forward flight airspeeds implied, of such products or manufacturers by the National Aeronautics and for the RAVEN-SWFT vehicle are higher than the airspeed Space Administration.

subsystem is configured to allow for real-time piloted simula- Motor Performance The vehicle motor performance is tions using a USB transmitter for pilot training and assessment modeled using RSEs that convert motor rotational speed, col- of vehicle handling qualities, as well as accelerated automated lective pitch angle, nacelle tilt angle, and body-axis velocity simulations with input injections for FCS evaluation across to motor power consumption. The RSEs were identified using the flight envelope. The subsystem is designed to closely emu- data collected from the isolated proprotor and powered air- late the input functionality and structure for the FCS deployed frame wind-tunnel testing described previously. Although the onto the vehicle. motor performance models do not influence the vehicle dy- namics in the simulation, they help to inform trim strategies and flight-test planning.

Simulation Outputs The Simulation Outputs subsystem stores, plots, and creates a Aero-Propulsive Model The aero-propulsive model is com- visualization of the simulation data. The data from the other posed of the RSEs identified from wind-tunnel testing, as de- subsystems are routed to simulation scopes and stored in the scribed in the previous section (Refs. 15, 16). The RSEs per- ® MATLAB workspace for use by post-processing scripts to taining to specific dynamic pressure settings are blended using facilitate desktop evaluation of the FCS performance and sim- shape-preserving piecewise cubic interpolation (Refs. 52, 53) ulation testing to meet software assurance requirements.

to form a continuous transition model. After computing the The vehicle motion predicted by the flight dynamics simula- aero-propulsive model predictions as a function of state and tion can be visualized in real time using a simple generic UAV control variables, the moment predictions are transferred from ® Animation from the MathWorks UAV toolbox (Ref. 49) the wind-tunnel moment reference center location to the vehi- or using a high-fidelity RAVEN-SWFT visualization imple- cle center of gravity location.

mented in X-Plane (Ref. 50), along with a toggleable real-time simulation pacer. The custom RAVEN-SWFT X-Plane vi- Equations of Motion The equations of motion subsystem sualization includes accurate vehicle geometry and animated includes the kinematic and dynamic aircraft equations of mo- control effectors, which is helpful for pilot training, mission tion developed under a standard set of assumptions (Refs. 54– rehearsals, and qualitative FCS evaluation. The X-Plane visu- 56). The aircraft is nominally modeled as a single six-degree- alization was developed and implemented following the steps of-freedom (6DOF) rigid body subjected to gravitational force given in Ref. 51.

and aero-propulsive forces and moments. The dynamics asso- ciated with the rotating portion of each propulsor and nacelle Vehicle Model tilt are currently neglected since the inertia of rotating vehicle components is small compared to the overall vehicle inertia; The Vehicle Model subsystem includes a control effector dy- however, multi-body equations of motion (e.g., see Ref. 57) namics subsystem, a motor performance subsystem, an aero- are planned to be added to the nonlinear simulation in the propulsive model subsystem, an equations of motion subsys- future. The simulation states are the body-axis translational tem, a sensors subsystem, a ground contact subsystem, and an velocity ( 𝑢, 𝑣, 𝑤 ), body-axis angular velocity ( 𝑝, 𝑞, 𝑟 ), attitude atmospheric wind/turbulence subsystem. The contents of the quaternions ( 𝑞 , 𝑞 , 𝑞 , 𝑞 ), and earth-fixed position in terms 0 𝑥 𝑦 𝑧 vehicle simulation subsystems are highlighted in the follow- of latitude, longitude, and altitude ( Φ , Ψ , ℎ ). Because the sim- ing paragraphs.

ulation was used to design a flight controller for a 3DOF wind- tunnel test, in addition to free flight testing, the 6DOF simu- Control Effector Dynamics The control effector dynamic lation was programmed to include necessary modifications to models describe the control surface servo-actuator position conduct 3DOF simulations (Ref. 16).

and propulsor rotational speed response subject to changes in the commanded value. Control surface actuation is modeled Sensor Model The simulated sensor data were optionally using a first-order dynamic model, whereas the propulsor ro- corrupted with white, Gaussian measurement noise. The sen- tational speed dynamics are described using a second-order sor model also includes a bias, scale factor, and time delay.

dynamic model. The first- and second-order dynamic models The noise properties were determined based on experimental also include a rate limit, time delay, and position limit satu- data collected for the RAVEN-SWFT vehicle. The structure ration. Including the rate limit and saturation introduces non- of the sensor subsystem closely follows the implementation in linearity into the control surface and propulsor dynamic mod- Ref. 48.

els. Although linear dynamic models are more convenient for control system design and evaluation, including the rate limit and saturation in the model provides a more accurate repre- Ground Contact Model A ground contact model is included sentation of the actual actuator dynamics in the high-fidelity to allow simulation of takeoff and landing operations for soft- nonlinear flight dynamics simulation. The control surface and ware verification and pilot training. The current simplified propulsor dynamic model parameters were determined using ground contact model is not intended to accurately represent data collected from step inputs applied to each individual con- the contact forces and moments of the vehicle landing gear trol effector under aerodynamic loading in wind-tunnel test- with the ground, but rather is designed to emulate the visual ing. behavior of takeoff/landing operations from a remote pilot perspective. An accurate physics-based ground contact model steady, level, unaccelerated flight at different airspeed condi- is planned to be added in the future. tions throughout the transition envelope. To improve the oper- ational practicality and convergence of the trim optimization algorithm, certain control effectors were strategically linked Atmospheric Model The atmospheric model is identical while formulating the optimization problem. The free and/or to subsystems included in Ref. 48, which include the first fixed nonzero variables in the optimization algorithm were: layer of the 1976 U.S. Standard Atmosphere model (Ref. 58), • Pitch angle, 𝜃 (angle of attack, 𝛼 ) steady winds, and the Dryden turbulence model (Refs. 59,60).

• Front proprotor rotational speed, 𝑛 = 𝑛 = 𝑛 = 𝑛 = 𝑛 front 1 2 3 4 Because the RAVEN-SWFT always operates close to sea level • Rear proprotor rotational, 𝑛 = 𝑛 = 𝑛 rear 5 6 conditions, as currently configured, the aero-propulsive model • Front collective pitch angle, 𝛿 = 𝛿 = 𝛿 = 𝛿 = 𝛿 𝑐 𝑐 𝑐 𝑐 𝑐 assumes constant air properties at sea level conditions. front 1 2 3 4 • Rear collective pitch angle, 𝛿 = 𝛿 = 𝛿 𝑐 𝑐 𝑐 rear 5 6 • Nacelle tilt angle, 𝛿 = 𝛿 = 𝛿 = 𝛿 = 𝛿 (except in 𝑡 𝑡 𝑡 𝑡 𝑡 1 2 3 4 FLIGHT CONTROL SYSTEM DESIGN hover, where nacelle tilt angle was split into outboard nacelle tilt angle 𝛿 = 𝛿 = 𝛿 and inboard nacelle tilt 𝑡 𝑡 𝑡 out 1 4 The flight control algorithm was developed with a model- angle 𝛿 = 𝛿 = 𝛿 ) 𝑡 𝑡 𝑡 in 2 3 based design approach, using the RAVEN-SWFT flight dy- • Outboard flap deflection angle, 𝛿 = 𝛿 = 𝛿 𝑓 𝑓 𝑓 out 1 6 namics simulation. The model-based approach allows for • Inboard flap deflection angle, 𝛿 = 𝛿 = 𝛿 = 𝛿 = 𝛿 𝑓 𝑓 𝑓 𝑓 𝑓 in 2 3 4 5 specification of performance and robustness requirements, use • Stabilator deflection angle, 𝛿 𝑠 of optimal control techniques for gain tuning, rapid develop- Figure 7 shows the free and fixed variable definitions at each ment in a simulation environment, and increased confidence flight condition.

in the flight control solutions for vehicle testing (Ref. 61).

The trim optimization objective function was the total mo- The overall RAVEN-SWFT flight control algorithm is com- tor power usage added to the sum of each motor rotational posed of a nested-loop architecture, depicted in Fig. 5, includ- speed and collective pitch angle in coded units, formed from ing an attitude-control inner loop and a velocity-control outer the operational bounds, raised to an even power. The objective loop. The design approach includes a linear quadratic inte- function balances minimizing power consumption (i.e., max- gral (LQI) (Ref. 62) model-following control framework with imizing vehicle range) and maximizing the available propul- a static feedforward element. The nested-loop architecture al- sion control authority. Figure 8 provides a visualization of lows for sequential design of the flight control laws and mul- the trim settings for the pitch attitude ( 𝜃 ), motor rotational tiple entry points for pilot commands, facilitating a build-up speed ( 𝑛 ), collective pitch angle ( 𝛿 ), nacelle tilt angle ( 𝛿 ), 𝑐 𝑡 approach to flight-test verification. The inner- and outer-loop flap deflection ( 𝛿 ), and stabilator deflection ( 𝛿 ). The black 𝑓 𝑠 control laws have the same overall structure, shown in Fig. 6.

dotted vertical line at 65 ft/s on each plot represents the limits Each has filtered sensor data used for feedback stabilization of the high-fidelity wind-tunnel database. As mentioned pre- and reference tracking, a command model for specifying the viously, the high-speed transition and forward flight airspeeds desired system response to reference inputs, a static feedfor- for the RAVEN-SWFT vehicle are higher than were able to be ward path for transient response shaping, and a control allo- tested in the 12-Foot LST; accordingly, the high-speed transi- cation scheme to appropriately distribute the control system tion model at airspeeds greater than 65 ft/s is lower-fidelity, commands.

which increases uncertainty in the trim solution. The trim The following subsections describe the current RAVEN- solutions are expected to be updated after conducting flight SWFT trim approach, control allocation, control law design, testing at these high-speed conditions.

control modes, and takeoff/landing sequence, as of April 2025. The architecture, parameters, and implementation of Control Allocation the RAVEN-SWFT FCS have continuously evolved as new data have been obtained from wind-tunnel and flight tests. A weighted pseudo-inverse control allocation approach was Therefore, it is expected that the FCS will continue to be up- used to determine the control effector commands 𝒖 to cmd dated with the collection of new flight-test data and aspects of achieve desired force and moment commands 𝝁 : the control design approach described herein may change in   − 1 T T − 1 − 1 ¯ ¯ ¯ the future.

𝒖 = 𝑾 𝑩 𝑩𝑾 𝑩 𝝁 (1) cmd ¯ Here, 𝑾 is a diagonal weighting matrix and 𝑩 is the control Trim Approach effectiveness matrix. Following the approach of Ref. 63, the nominal diagonal element weight 𝑤 assigned to the 𝑖 th effec- 𝑖 The design of the flight control algorithm is based on lin- tor was selected using its rate limit 𝑅𝐿 as: 𝑖 earization of the nonlinear flight dynamics simulation at trim conditions. Trim settings, where the sum of gravitational 𝑤 = (2) and aero-propulsive forces and moments are equal to zero, 𝑖 𝑅𝐿 𝑖 were determined across the flight envelope using the con- ¯ strained nonlinear optimization function fmincon available The 𝑩 matrix for the inner-loop allocation included only rows ® ¯ in MATLAB (Ref. 49). Trim analysis was performed for corresponding to the rotational dynamics equations. The 𝑩 Figure 5: Schematic of the overall RAVEN-SWFT flight control algorithm.

Figure 6: Schematic of the inner- and outer-loop control law structure.

cies, including effectors that have a secondary effect on an axis being over used, sharp changes in the allocation between control design points across the envelope, and adverse inter- axis coupling. Modifications to the standard weighted-pseudo inverse allocation solution were implemented to improve the overall allocation strategy. One modification was to set the ¯ elements of the 𝑩 matrix that did not contribute the the dom- inant control effects for the particular axis to zero. Addition- ally, certain nominal weights 𝑤 at different control design 𝑖 points were multiplied by a scale factor that was adjusted us- ing a constrained nonlinear optimization algorithm with a cost function designed to minimize the variation in each element of the allocation matrix between control design points. Both of Figure 7: Definition of free and fixed trim optimization vari- these modifications result in a more practical allocation solu- ables at each flight condition.

tion, improving the overall flight controller performance. In- vestigating alternative control allocation strategies using lin- matrix for the outer-loop allocation included rows correspond- ear or quadratic programming (Ref. 64) or cascading general- ing to the translational and rotational dynamics equations; ized inverse (Ref. 65) allocation methods is a planned area of however, only the columns of the allocation matrix from the future research.

weighted pseudo-inverse solution that correspond to the trans- lational dynamics equations were used for the outer-loop con- Flight Control Law Design trol allocation. The outer-loop allocation also included two ¯ additional columns appended to the end of the 𝑩 matrix for As mentioned previously, the RAVEN-SWFT FCS utilizes the roll and pitch command channels, with weights set as an nested inner- and outer-loop flight control laws, shown in approximation of the rate limit for roll and pitch attitude com- Figs. 5-6, to control the rotational and translational motion mands.

of the aircraft, respectively. The inner- and outer-loop flight Initial simulations using the conventional weighted pseudo- control laws each implement a model-following control ap- inverse control allocation strategy revealed practical deficien- proach with LQI feedback and a static feedforward inverse that arises from the solution to the minimization of the quadratic performance index ∫ ∞ T T 𝐽 = 𝒙 𝑸𝒙 + 𝒖 𝑹𝒖 (8) 𝑎 𝑎 where 𝑸 ≥ 0 and 𝑹 > 0 . The optimal feedback gain is then − 1 T 𝑲 = 𝑹 𝑩 𝑷 (9) 𝑎 which can be divided into integral and state feedback gain as   𝑲 𝑲 𝑲 = (10) 𝑖 𝑥 Defining 𝑞 as the diagonal elements of 𝑸 , tuning was ac- 𝑖𝑖 complished via a line search over the elements corresponding to the integral states at each trim point, with manual tuning of the elements corresponding to the system states. The in- put matrix weight was set as 𝑹 = I for all trim points. The full 4-state models were used for the design of the inner-loop control law, with pitch attitude tracking in the longitudinal channel and bank angle and yaw rate (defined in either the body or heading frame) in the lateral-directional channel. The outer-loop design also uses the 4-state linear models, where the velocity states can be defined in either the body frame or heading frame, with heading frame longitudinal and vertical speed command tracking in the longitudinal channel and lat- eral speed tracking in the lateral channel.

The longitudinal and lateral-directional control laws in both the inner and outer loops have an identical structure, depicted in Fig. 6. The outer-loop control law allocation sends com- Figure 8: Variation of RAVEN-SWFT trim settings with mands to the aircraft actuators and to the input of the inner- freestream airspeed.

loop, which can be used to specify desired behaviors (e.g., coordinated turns in response to a heading change command).

model approximation.

Command Model and Inverse Model Design The com- LQI Feedback Design To design the LQI feedback gains, mand model defines the desired response of the system to an linear models were obtained from the nonlinear simulation input command and is specified based on response type. The at each of the trim conditions. The linear models were then inner-loop control law uses a second-order response type for decomposed into 4-state longitudinal ( 𝑢, 𝑤, 𝑞, 𝜃 ) and lateral- all channels defined by directional ( 𝑣, 𝑝, 𝑟, 𝜙 ) systems, where the velocity states can be defined in either the body frame ( 𝑢 , 𝑣 , 𝑤 ) or heading frame 2 𝑦 𝜔 CM 𝑛 = (11) ( ¯ 𝑢 , ¯ 𝑣 , ¯ 𝑤 ) (Refs. 19, 20). The linear systems are then of the 𝑟 𝑠 + 2 𝜁𝜔 + 𝜔 𝑛 𝑛 form while the outer-loop control law implemented a first-order re- ¤ 𝒙 = 𝑨𝒙 + 𝑩𝒖 (3) sponse type: 𝒚 = 𝑪𝒙 (4) 𝑦 𝜔 CM = (12) 𝑟 𝑠 + 𝜔 Then, defining 𝒆 = 𝒚 − 𝒓 , the augmented system The design of the command models accounted for desired ¤ 𝒙 = 𝑨 𝒙 + 𝑩 𝒖 + 𝑩 𝒓 (5) 𝑎 𝑎 𝑎 𝑎 𝑟 open-loop rise time, settling time, and actuator rates and de- flections for a nominal input command. A static approxima- 𝒚 = 𝑪 𝒙 (6) 𝑎 𝑎 tion of the open-loop inverse dynamics is computed to pro- can be used for the LQI design, where vide a feedforward command that ensures the open-loop air-       craft response matches the command model response. Fol-   0 𝑪 0 − I 𝑨 = , 𝑩 = , 𝑩 = , 𝑪 = 0 𝑪 lowing the approach in Ref. 66, a low-order equivalent system 𝑎 𝑎 𝑟 𝑎 0 𝑨 𝑩 0 is fit to the open-loop response (with stabilizing state feedback 𝒖 = − 𝑲 𝒙 ) of the form 𝑥 Letting 𝒖 = − 𝑲𝒙 , the optimal feedback gain, 𝑲 , is computed 𝑎 from the solution, 𝑷 , to the algebraic Ricatti equation 2 − 𝜏 𝑠 inv 𝐾 𝜔 𝑒 𝑦 inv inv = (13) T − 1 T 2 2 𝜇 𝑷 𝑨 + 𝑨 𝑷 − 𝑷𝑩 𝑹 𝑩 𝑷 + 𝑸 = 0 (7) 𝑠 + 2 𝜁 𝜔 𝑠 + 𝜔 𝑎 𝑎 inv inv 𝑒 𝑎 inv for the second-order command model and command variable violates the time-scale separation assump- tion and introduces undesired coupling between the scheduled − 𝜏 𝑠 inv 𝑦 𝐾 𝜔 𝑒 inv inv = (14) state and the control system, precluding the use of measured 𝜇 𝑠 + 𝜔 inv forward velocity as a scheduling variable. The gain schedul- ing approach chosen for RAVEN-SWFT allows for a full- for the first-order command model. Then, the feedforward envelope flight controller that transitions the aircraft between command term is defined as hover and forward flight conditions.

𝑢 = 𝐾 𝑟 + 𝐾 𝑦 + 𝐾 ¤ 𝑦 (15) ff ff 𝑝 CM 𝑑 CM CM CM Feedback Filters VTOL vehicles generally experience sig- where 𝑦 is the output of the command model and ¤ 𝑦 is CM CM nificant vibrations due to the proprotor rotation and excited the time derivative of the command model output. For the structural resonance. To reduce high-frequency vibrations second-order command model: from propagating into the feedback signals, a series of low- 2 2 𝐾 = 𝜔 /( 𝐾 𝜔 ) pass and notch filters were placed on the feedback signals up- ff inv 𝑛 inv stream of the feedback control laws. The filter parameters 2 2 2 𝐾 = ( 𝜔 − 𝜔 )/( 𝐾 𝜔 ) 𝑝 𝑛 inv CM inv inv were set based on flight data and were designed to sufficiently 𝐾 = ( 2 𝜁 𝜔 − 2 𝜁𝜔 )/( 𝐾 𝜔 ) 𝑑 inv inv 𝑛 inv CM attenuate undesired high-frequency content while also mini- inv mizing the delay added to the feedback signals by the filters.

For the first-order command model: 𝐾 = 𝜔 /( 𝐾 𝜔 ) ff inv inv Flight Control Modes 𝐾 = ( 𝜔 − 𝜔 )/( 𝐾 𝜔 ) 𝑝 inv inv inv CM The RAVEN-SWFT FCS includes three separate FCS modes: 𝐾 = 0 𝑑 CM FCS Mode 0, FCS Mode 1, and FCS Mode 2. Mode 0 in- cludes inner-loop attitude stabilization and direct thrust com- The time delay 𝜏 = 𝜏 is used to delay the reference com- cmd inv mands to the propulsors; Mode 0 is only intended for hover mand to the LQI input, as used in Ref. 66 and shown in Fig. 6.

and primarily exists to reduce risk in initial flight testing.

Computing the feedforward gains in this manner provides a Mode 1 includes inner-loop attitude stabilization with an static-gain approximation of an inverse model of the dynam- outer-loop altitude rate controller. Mode 2 is a full unified- ics for each channel, resulting in simpler design and imple- velocity controller with an attitude-control inner loop and mentation than using a transfer function approximation for the a velocity-control outer loop; Mode 2 includes the highest inverse model with comparable performance. The command amount of augmentation and reflects a simplified vehicle op- models are implemented in the flight controller as state-space erations approach that has parallels to pilot command ap- systems, so that the command model derivative, ¤ 𝑦 is di- CM proaches pursued for passenger-carrying eVTOL aircraft de- rectly accessible without numeric differentiation.

signs (Ref. 67).

Tuning Approach Control system tuning was performed in The pilot RC transmitter command variables change based on a sequential and iterative manner. The inner-loop tuning pro- the FCS mode, and are summarized in Table 1. In Mode 0 cess starts from the control allocation and works outwards to and Mode 1, the right stick commands pitch and roll attitude, the LQI control law, command model and feedforward ele- whereas in Mode 2 the right stick commands heading-frame ment. Then, the outer-loop tuning progress proceeds simi- forward and lateral velocity changes. Side-to-side left stick larly through the control allocation, LQI control law, com- movement commands heading-frame yaw rate in each con- mand model, and feedforward element.

trol mode. Up-down left stick movement commands thrust changes in Mode 0, whereas it commands vertical velocity in Control performance and robustness analyses were performed Mode 1 and Mode 2. In each FCS mode, the left slider com- using both the linearized and nonlinear flight dynamics mod- mands the forward velocity setting from zero to the maximum els. Control law synthesis accounted for closed-loop perfor- specified forward velocity, which results in nacelle tilt angle mance including rise time, settling time, and response over- changes to transition the aircraft.

shoot/undershoot; crossover frequency; actuator usage (mag- nitude and rates); and gain and time-delay margins at the plant input in the nonlinear dynamics simulation.

Takeoff and Landing Sequence Gain Scheduling Control law gains, control allocation, trim Key phases of flight for an eVTOL vehicle include its hov- settings, and other FCS parameters, are scheduled with nom- ering takeoff and landing. The piloted takeoff sequence was inal nacelle tilt angle. The commanded nacelle tilt angle is given particular focus for the first phase of RAVEN-SWFT scheduled with commanded heading-frame forward velocity flight testing. The pilot uses switches and sticks on the RC and is determined in flight by a rate-limited forward velocity transmitter to execute the takeoff sequence. The takeoff se- command. This approach was chosen over more conventional quence follows a list of pre-flight checklist items to ensure direct airspeed scheduling since forward velocity is a com- personnel and vehicle safety. The current piloted takeoff se- mand variable. Scheduling on a vehicle state that is also a quence, from the flight control perspective, is as follows: Table 1: Pilot command variable definitions for each RAVEN-SWFT flight control mode Pilot Command Mode 0 Mode 1 Mode 2 Interface Type (Direct Thrust) (Vertical Velocity) (Unified Velocity) Right Stick ( ↔ ) Lateral Roll Attitude ( 𝜙 ) Roll Attitude ( 𝜙 ) Lateral Velocity ( ¯ 𝑣 ) Right Stick ( ↕ ) Longitudinal Pitch Attitude ( Δ 𝜃 ) Pitch Attitude ( Δ 𝜃 ) Forward Velocity ( Δ ¯ 𝑢 ) ¤ ¤ ¤ Left Stick ( ↔ ) Directional Yaw Rate ( 𝜓 ) Yaw Rate ( 𝜓 ) Yaw Rate ( 𝜓 ) Left Stick ( ↕ ) Vertical Thrust ( Δ 𝑇 ) Vertical Velocity ( ¯ 𝑤 ) Vertical Velocity ( ¯ 𝑤 ) Left Slider Transition Forward Velocity ( ¯ 𝑢 ) Forward Velocity ( ¯ 𝑢 ) Forward Velocity ( ¯ 𝑢 ) 1. Ensure that the two arm switches are in the upward “dis- to determine when the vehicle has landed and automatically arm” (0) position. disarms the vehicle. Increasing the takeoff and landing au- 2. Select the desired flight control mode using the FCS tomation, as well as the automated flight missions using the Mode switch on the RC transmitter. RAVEN-SWFT custom flight control software, is planned for 3. While holding the vertical stick command to its nega- future flight control upgrades.

tive limit (full downward left stick), move one of the PROGRAMMED TEST INPUTS arm switches to the center “arm” (1) position, and then release the self-centering vertical stick. This will arm One key feature of the custom RAVEN-SWFT FCS is a pro- the vehicle in QGroundControl and allow the the pilot to grammed test input (PTI) injection capability used for sys- move all of the control effectors. The proprotor speed tem identification. The PTI excitations are summed with the will remain at zero for this step.

reference command and control effector signals, before and 4. Deflect each of the control sticks and visually ensure that after the flight control laws, as shown in Fig. 9. The auto- each of the 18 servos are active and properly command- mated input types included multistep (Refs. 54,68), frequency ing control effector position.

sweep (Refs. 54, 69), and multisine excitations (Ref. 54), 5. Move the second arm switch to the center “arm” (1) po- which were created using the System IDentification Programs sition (i.e., place both arm switches in the center posi- for AirCraft (SIDPAC) software toolbox (Ref. 70).

tion). The servos will deflect their corresponding control effector to the trim hover setting. Then, after a brief de- Control Effector Reference Command lay to allow the servos to reach their hover positions, the PTI Excitations PTI Excitations proprotors will start spinning at the minimum rotational speed setting. Visually ensure that each of the proprotors Control Reference Control Effector + + + + Inputs Commands Laws is spinning.

6. Move both arm switches to the downward “engage” (2) Sensor Data position. Then, move the vertical stick down to its most negative position. This will engage FCS Mode 0 at the Figure 9: PTI injections relative to the control laws.

minimum thrust setting with the integrators disabled to prevent integrator windup.

Square wave multistep inputs (e.g., doublet, 1-2-1, and 3-2- 7. Carefully move the vertical stick towards the center po- 1-1 inputs) and frequency sweep inputs are injected into the sition, which will increase the proprotor speed and col- reference command PTIs upstream of the control laws. Cur- lective pitch angle towards the trimmed hover thrust set- rently, doublet (1-1) commands are used to evaluate the con- tings. The vehicle will elevate above the ground near the trol system tracking performance and serve as system iden- center vertical stick position.

tification validation maneuvers. Frequency sweep and the 8. Once the vehicle’s distance above ground level surpasses more complex multistep inputs have been exercised for ver- a predefined small separation distance from the ground, ification purposes, but are not currently planned to be used in as sensed by a LiDAR distance sensor, the FCS will the RAVEN-SWFT flight testing. Multisine inputs, described switch into the desired flight control mode set above and next, are the preferred input type for efficient open-loop model enable the integrators in the control laws. This completes identification for the present application due to the large num- the takeoff sequence and the pilot can then execute the ber of control effectors and aero-propulsive complexity asso- flight mission designated for the particular flight.

ciated with eVTOL vehicles.

The current piloted landing sequence is simply to guide the Orthogonal phase-optimized multisine inputs (Refs. 54, 71– vehicle to a gentle landing and then promptly move both arm 73) are the primary PTI type used for system identification switches to the upward “disarm” (0) position, which will set for the RAVEN-SWFT aircraft. A multisine input is de- the proprotor speed commands to zero. The pilot can then fined as a sum of multiple sinusoidal functions with differ- proceed through the takeoff sequence again to perform an- ent amplitudes, frequencies, and phase angles, where the fre- other takeoff. An additional automated landing sequence has quencies are chosen to encompass the frequency range cor- been successfully tested that uses the LiDAR distance sensor responding to the system dynamics of interest. To make all inputs orthogonal in both the time domain and frequency do- 12, and 5 frequency components, respectively. The frequency main, the multisine signal for the 𝑗 th control effector is as- components were assigned to each signal in an alternating signed sinusoids with a unique subset 𝐾 of discrete har- manner and were distributed to avoid neighboring control ef- 𝑗 monic frequency indices selected from the complete set of fectors having neighboring frequency components, similar to 𝐾 available frequency indices. The available frequencies are the approach described in Ref. 74.

𝑓 = 𝑘 / 𝑇, 𝑘 = 1 , 2 , ..., 𝐾 , where 𝑇 is the fundamental period 𝑘 The input spectra for the final set of RAVEN-SWFT flight- and 𝐾 / 𝑇 is the highest excitation frequency. For 𝑚 total con- test orthogonal phase-optimized multisine signals is shown trol effectors, the 𝑗 th input signal 𝑢 ( 𝑡 ) is defined as 𝑗 in Fig. 10. There are 428 total harmonic components with   a fundamental period of 𝑇 = 260 seconds. The overall fre- ∑︁ √︁ 2 𝜋𝑘𝑡 𝑢 ( 𝑡 ) = 𝐴 𝑃 sin + 𝜙 , 𝑗 = 1 , 2 , ..., 𝑚 (16) quency range is between 𝑓 = 0 . 05 Hz and 𝑓 = 1 . 754 Hz 𝑗 𝑗 𝑘 𝑘 min max 𝑇 𝑘 ∈ 𝐾 with a frequency resolution of Δ 𝑓 = 1 / 𝑇 = 0 . 00385 Hz. Fig- 𝑗 ure 11 shows the first 20 seconds of the input excitation sig- where 𝐴 is the signal amplitude, 𝑃 is the 𝑘 th power fraction 𝑗 𝑘 nals with 𝐴 = 1 . For system identification experiments, a Í 𝑗 (with 𝑃 = 1 ), 𝜙 is the 𝑘 th phase angle defined on the 𝑘 ∈ 𝐾 𝑘 𝑘 𝑗 gain is applied to scale each input signal to a sufficient ampli- interval (− 𝜋, + 𝜋 ] , and 𝑡 is the elapsed time. The phase an- tude to obtain a good signal-to-noise ratio, while not deviating gles are optimized to obtain a minimum relative peak factor, far from the trimmed flight condition. Note that the nomi- thereby helping to keep the system close to its nominal flight nal RAVEN-SWFT 60-second system identification maneu- trajectory while also retaining the same high input energy as a ver length is longer than the 20-second input signal duration signal without optimized phase angles. Because multiple in- shown for demonstration purposes in Fig. 11.

puts and all aircraft dynamics of interest are simultaneously Because the RAVEN-SWFT is unstable over a significant por- excited, the use of multisine inputs allows execution of highly tion of its flight envelope, the feedback control system must be efficient and informative flight testing. A single multisine active when operating the aircraft. Therefore, the control ef- maneuver can be used to develop a comprehensive aircraft fector signals sent to the motors and actuators in flight testing dynamic model around a nominal flight condition, including are different than the designed signals shown in Fig. 11 and nonlinear aerodynamic phenomena and control interaction ef- become somewhat correlated due to distortion from the active fects.

FCS; however even with this distortion and diminished or- For RAVEN-SWFT flight testing, individual multisine signals thogonality, the modeling variables are generally sufficiently were generated for each of the 24 independent control effec- decorrelated for model identification because the control ef- tors. Additionally, multisine signals were created for the ref- fector excitation inputs are added downstream of the flight erence commands ( 𝜂 , 𝜂 , 𝜂 , 𝜂 ) tracked by the control lon lat dir ver control laws (Refs. 54, 73). An active control system may system to provide additional state perturbations supplement- still slightly degrade modeling results, as compared to mod- ing state disturbances resulting from control effector PTI in- eling a vehicle that does not require a control system to be jections. In previous RAVEN-SWFT 3DOF wind-tunnel test- active while executing system identification maneuvers, but ing, the multisine directional reference commands were noted the adverse effects of the FCS on the system identification ex- to be particularly helpful for improving the parameter estimate periment are largely mitigated by following this approach.

values associated with lateral speed 𝑣 and yaw rate 𝑟 (Ref. 16).

To execute an automated RAVEN-SWFT system identifica- Combining the control effector and reference command multi- tion maneuver, starting from a trimmed flight condition, the sine signals, which are all executed simultaneously, there are pilot reverses the polarity of a PTI slider on the RC trans- a total of 28 independent excitation signals. The RAVEN- mitter to enable the PTI excitations. The vehicle FCS then SWFT flight-test multisine design has some minor differences injects the selected PTI into the control effector and/or refer- compared to the 27 component RAVEN-SWFT multisine de- ence commands, perturbing the vehicle around its reference sign described in Ref. 16 that was used for RAVEN-SWFT conditions without requiring any additional pilot inputs. Af- 3DOF wind-tunnel testing.

ter completing the maneuver, the pilot returns the PTI slider To create the RAVEN-SWFT multisine signals used for flight to its original position to disable the PTI excitations.

testing, several harmonic components were assigned to each The form of the PTI injected by the FCS is governed by mul- multisine signal, where the overall frequency range was set tiple RC transmitter channels. The “PTI Type” (multisine, to between 0.05 Hz and 1.75 Hz in accordance with frequen- frequency sweep, or multistep) is set using a three position cies where the rigid-body dynamics were predicted to mani- switch on the RC transmitter. The “PTI Option” for each PTI fest. The harmonic components for the reference commands Type is then able to be selected using a combination of two and slower moving control effectors were selected to be con- additional three position switches, called the “PTI Mode” and sistent with the approximate bandwidth of each signal. The “PTI Submode” as: collective pitch, outboard tilt, stabilator, flaperon, and rud- der multisine signal harmonic components spanned the full [ PTI Option ] = 3 × [ PTI Mode ] + [ PTI Submode ] frequency range and were each assigned 18 frequency com- ponents. The proprotor rotational speed, inboard tilt, and atti- Table 2 lists the current PTI Option settings for each PTI Type, tude command harmonic components were focused into lower where there is room for future expansion up to 9 options for frequencies due to their lower bandwidth and contained 16, each PTI Type.

Figure 10: RAVEN-SWFT flight-test multisine input spectra for each reference command and control effector.

Figure 11: Normalized RAVEN-SWFT multisine input signal for each reference command and control effector.

multistep inputs, the square wave step type (e.g., 1-1, 1-2-1, Table 2: RAVEN-SWFT PTI option descriptions 3-2-1-1) and step time interval can be adjusted using parame- ters set on the GCS. The current RC transmitter interface and PTI GCS PTI parameters are discussed further in the next section.

PTI Type Option PTI Functionality Multisine 0 Control Effector (CE) Multisine 1 Reference Cmd. (RC) Multisine CONTROL SYSTEM INTEGRATION 2 Combined CE and RC Multisine In parallel to simulation-based RAVEN-SWFT FCS develop- Frequency 0 Lateral Command Sweep ment, efforts were undertaken to identify a streamlined so- Sweep 1 Longitudinal Command Sweep ® lution to deploy a custom Simulink model onto flight ve- 2 Directional Command Sweep hicle hardware. Through the use of low-cost surrogate vehi- 3 Vertical Command Sweep cles, a process to deploy custom flight control logic developed Multistep 0 Lateral Command Steps ® in Simulink onto a Pixhawk flight computer was identified 1 Longitudinal Command Steps and refined (Refs. 51, 75, 76). The flight controller deploy- 2 Directional Command Steps ® ment process, which uses the MathWorks UAV Toolbox and 3 Vertical Command Steps its Support Package for PX4 Autopilots (Ref. 49), was de- termined to be a suitable solution for RAVEN-SWFT with a To achieve an appropriate amount of vehicle excitation, the few vehicle hardware-related modifications. The default PX4 overall PTI amplitude is scaled using a rotary knob on the outer- and inner-loop control laws are replaced by the custom RC transmitter. Furthermore, the individual command multi- RAVEN-SWFT control laws described above, but the flight step/frequency sweep amplitude, as well as the multisine am- controller deployed onto the Pixhawk retains most of the other plitude for groups of similar control effectors (e.g., outboard functionality in the PX4 Autopilot firmware, such as the state flaperons, inboard flaperons, front collective, rear collective, estimation algorithm, input/output drivers, safety logic, and etc.) and each reference command, can be scaled separately by data logging capabilities. These methods allow for rapid de- modifying parameters on a GCS computer. Additionally, for ployment of a custom flight control algorithm, where a new control software update can be built and deployed onto the • Proprotor rotational speed Pixhawk flight computer in under five minutes. • Above ground distance The vehicle attitude, angular velocity, and inertial velocity are After the RAVEN-SWFT FCS was designed and tested in sim- estimated states obtained from either the PX4 state estimation ulation, steps were taken to integrate the control logic onto the algorithm or an external inertial navigation system (INS). The vehicle hardware. Initially, simulations were executed with ® external INS was found to improve FCS performance and is the Simulink model running on a laptop computer and com- the default state estimation source for RAVEN-SWFT. Inertial municating with the vehicle sensors, control effectors, and ® position and winds are from the PX4 state estimator. True air- RC transmitter/receiver. Then, the Simulink model was au- speed, proprotor rotational speed, and above ground distance tocoded onto a Pixhawk flight computer to execute experi- are from an external airspeed sensor, the proprotor ESCs, and mental testing. The top level of the flight control software ® an external LiDAR sensor, respectively.

Simulink diagram used for integration onto the RAVEN- SWFT vehicle is shown in Fig. 12, which includes the in- RC Transmitter An RC transmitter capable of commanding puts from the vehicle, flight control system, and outputs to 16 independent channels is used as the pilot interface to com- the vehicle. The overall structure is similar to the simulation mand the vehicle trajectory and execute various research func- structure shown in Fig. 4, except that the FCS Input and FCS tionality. The channels are composed of signals from the left Output subsystems in Fig. 12 are used to interface with the ve- and right stick position (4 channels), left and right sliders (2 hicle and the vehicle model is replaced by the actual vehicle channels), left and right rotary knobs (2 channels), and 8 dis- dynamics. The Flight Control System subsystem is identical crete switches (8 channels) including a two-position locking to the equivalent subsystem shown in Fig. 4; the subsystem switch and 7 three-position switches. Both of the pilot control contains the RAVEN-SWFT control algorithm and the PTI in- sticks are self centering due to the hands-off state-hold na- jection capabilities described previously, as well as other logic ture of the primary FCS modes (Modes 1 and 2). The assign- necessary to safely fly the vehicle.

ments of each transmitter channel to the corresponding FCS The following subsections describe the implementation of the response are summarized in Table 3. Additionally, Fig. 13 FCS onto RAVEN-SWFT hardware. The RAVEN-SWFT shows a schematic of the transmitter layout annotated with FCS described in this work was created and deployed us- the designated interface functionality. This transmitter con- ® ing MATLAB/Simulink R2024a (Ref. 49) and PX4 Version figuration builds on transmitter setup approaches discussed 1.14 (Ref. 77).

in Refs. 76 and 79 to meet the research requirements for the RAVEN-SWFT vehicle testing.

Table 3: RAVEN-SWFT RC transmitter channel descriptions for system identification flight testing Label Switch Type FCS Functionality P1 Right Stick ( ↔ ) Lateral Command Figure 12: Top-level view of the flight control software P2 Right Stick ( ↕ ) Longitudinal Command ® Simulink diagram integrated onto the vehicle.

P3 Left Stick ( ↔ ) Directional Command P4 Left Stick ( ↕ ) Vertical Command P5 Right Slider Enable/Disable PTI Flight Control System Inputs P6 Left Slider Transition ( ¯ 𝑢 ) Command P7 Right Knob Overall PTI Amplitude The FCS Inputs subsystem includes functionality to read sen- P8 Left Knob Miscellaneous Testing sor and state estimate data, read RC transmitter inputs, read SA 3-Position FCS Gain Adjustment GCS commands, and maintain an internal FCS elapsed time SB Locking Flight Termination counter.

SC 3-Position FCS Mode SD 3-Position Arm/Engage Switch 1 SE 3-Position Arm/Engage Switch 2 Vehicle Data Pertinent signals are read from the PX4 SF 3-Position PTI Type micro object request broker (uORB) communication sys- SG 3-Position PTI Mode tem (Ref. 78) using the UAV Toolbox “PX4 uORB Read” ® SH 3-Position PTI Submode Simulink block. The signals read into the RAVEN-SWFT FCS include: • Vehicle attitude in quaternions The RC transmitter sticks command the lateral, longitudinal, • Body-axis angular velocity directional, and vertical vehicle response based on the flight • Inertial velocity control mode (see Table 1). The flight control research com- • Latitude, longitude, and altitude mands are located on the left side of the transmitter, which in- • True airspeed cludes the FCS mode switch, a switch to enable FCS gain scal- • Estimated wind ing functionality, the transition forward velocity command on SB = Terminate SG = PTI Mode ↑ 0 = Terminate ↑ 0 = PTI Mode 0 – 1 = PTI Mode 1 P5 = Enable PTI P6 = ഥ 𝒖 Command ↓ 1 = Fly ↓ 2 = PTI Mode 2 ↑ = PTI Off ↑ = Max. ത 𝑢 SH = PTI Submode SA = FCS Gain Tuning ↓ = PTI On ↓ = Min. ത 𝑢 (0) ↑ 0 = PTI Submode 0 ↑ 0 = Original Gains Top of Transmitter – 1 = PTI Submode 1 – 1 = GCS Gain Scaling ↓ 2 = PTI Submode 2 ↓ 2 = Table Gain Scaling SC = FCS Mode SF = PTI Type SE = Arm 2 SD = Arm 1 P7 = PTI Amp. P8 = Misc .

↑ 0 = Direct Thrust ↑ 0 = Multisine ↑ 0 = Disarm ↑ 0 = Disarm – 1 = Vertical Vel. – 1 = Freq. Sweep – 1 = Arm – 1 = Arm ↓ 2 = Engage ↓ 2 = Unified Vel. ↓ 2 = Doublet ↓ 2 = Engage 0% 100% 0% 100% Upward Thrust/Vertical Velocity Pitch Down/Forward Velocity ↑ ↑ P1 P3 Left Yaw Rate ← O → Right Yaw Rate Left Roll/Velocity ← O → Right Roll/Velocity Front of Transmitter P2 P4 ↓ ↓ Downward Thrust/Vertical Velocity Pitch Up/Backward Velocity Figure 13: RAVEN-SWFT RC transmitter programming for modeling and controls research.

a slider, and a locking flight termination switch. The left ro- • A toggle between internal PX4 and external INS state tary knob is used to execute miscellaneous FCS tests (for ex- estimates (1 parameter) ample, the knob was temporarily programmed to initiate an • Distance sensor weight-on-wheels thresholds (2 parame- automatic landing command in initial flight testing). The sys- ters) tem identification research commands are located on the right • Feedback filter settings (5 parameters) side of the transmitter and include a slider to enable/disable • Servo-actuator calibration ground testing commands (4 PTIs, a PTI Type switch, a PTI Mode switch, a PTI Sub- parameters) mode switch, and a rotary knob assigned to adjust the over- • Flight control gain scaling values (15 parameters) all PTI amplitude. Furthermore, the two centermost switches • Forward velocity and acceleration limits (4 parameters) on the left and right side of the RC transmitter are used to- • Nacelle tilt command rate limits (2 parameters) gether to initiate the vehicle startup and shutdown sequence.

The custom system identification parameters included: Two switches are dedicated to this purpose to require two de- liberate actions by the pilot to initiate the arming sequence • Lateral, longitudinal, directional, and vertical reference (thereby, substantially reducing the likelihood of accidentally command multisine PTI amplitude (4 parameters) tripping this critical functionality). Similarly, the locking • Outboard flaperon, inboard flaperon, stabilator, rudder, flight termination switch requires two deliberate actions (lift- outboard nacelle tilt, inboard nacelle tilt, front collective, ing the switch and then changing its position) to initiate flight rear collective, front motor rotational speed, and rear mo- termination.

tor rotational speed multisine PTI amplitude (10 param- eters) In addition to the 16 RC channels, an additional PX4 parame- • Lateral, longitudinal, directional, and vertical reference ter is read to monitor the RC transmitter link and, if necessary, command multistep and frequency sweep PTI amplitude initiate the vehicle lost-link response. The lost-link response, (4 parameters) as well as other flight authorization features will be discussed • Lateral, longitudinal, directional, and vertical reference in the next section. The RC channels and transmitter link sig- command multistep PTI step time interval (4 parameters) nals are each read into the FCS using the UAV Toolbox “PX4 ® • Square wave multistep PTI waveform: doublet (1-1), 1- uORB Read” Simulink block.

2-1, or 3-2-1-1 (1 parameter) Ground Control Station Parameters Following the The GCS commands are individually read into the FCS using ® method described in Ref. 51, custom GCS parameters were the UAV Toolbox “PX4 Parameter Read” Simulink block.

created to adjust numerous flight control and system identifi- For safety considerations, the GCS parameters are only ad- cation parameters within the custom FCS on the GCS com- justed in QGroundControl (Ref. 80) while the vehicle is dis- puter. The custom flight control parameters included: armed on the ground; whereas, the RC transmitter is used to • Pilot stick sensitivity for reference commands (9 param- command the vehicle and execute/adjust research functional- eters) ity in real time while flying the vehicle.

Flight Control System Outputs status, proprotor rotational speed, FCS reference commands, FCS transition commands, control effector engineering com- The FCS Outputs subsystem sends commands to the control mands, sensor measurements and health status, and FCS/PTI effectors, transmits data to the GCS, and logs internal flight modes.

control signals.

Data Acquisition System Control Effector Commands The DroneCAN communica- In addition to the sensor, state estimate, and control data tion protocol (Ref. 81) is used for the communication with logged on the Pixhawk, a Data Acquisition System (DAS) was ESCs and control surface servo-actuators. Motor speed and developed to log data from the ESCs, control surface servo- control deflection commands from the FCS are converted to actuators, batteries, and an air-data probe. The DAS data used DroneCAN messages that are sent to the vehicle using the for post-test control and system identification analyses include ActuatorMotors and ActuatorServos uORB top- the commanded motor rotational speed, measured motor rota- ics accessed through the UAV Toolbox “PX4 uORB Write” ® tional speed, commanded servo position, actual servo posi- Simulink block. To accommodate the large number of con- tion, and air-data information. Additional data, such as the trol effectors present in eVTOL vehicle designs, a custom ver- ESC and servo voltage, current, and temperature, are useful sion of PX4 was created that included several modifications to for evaluating vehicle system performance.

allow the use of up to 20 CAN ESCs and 20 CAN servos.

The ESCs are sent commands that are directly related to the SOFTWARE ASSURANCE AND rotational speed requested by the flight controller. The ser- AIRWORTHINESS CONSIDERATIONS vos commands, however, are sent as scaled values between − 1 and + 1 that corresponds to the programmed minimum and A software assurance process was developed for the custom maximum position limits. Because the command to phys- RAVEN-SWFT FCS to meet software development require- ical deflection is generally a nonlinear relationship, a cali- ments and ensure safe flight operations, while also retain- bration curve was created and implemented into the FCS for ing agility in the development and modification of research each servo to convert between the physical commanded de- software for a subscale, modest cost vehicle. The RAVEN- flection requested by the FCS and servo commands sent to SWFT software development framework follows general en- the servos. The servo calibration is performed using a spe- gineering software best practices including version control, cial ground-based “servo calibration mode” built into the FCS software testing, detailed code documentation, and peer re- and executed using calibration-specific GCS parameters. A view. The RAVEN-SWFT simulation and FCS software is predetermined list of calibration commands in a randomized hosted and version controlled using an internal GitLab reposi- order are commanded through the GCS; the physical control tory (Ref. 83). The general process for modifying the software effector deflections are measured using a digital angle gauge is as follows: and recorded for each command. A polynomial is then fit to 1. Create an issue ticket describing the intended changes.

the calibration data to convert between command and physi- 2. Create a new branch from the main version of the soft- cal deflection, which is implemented into the FCS to compute ware remote repository associated with the issue ticket.

the CAN servo command.

3. Pull a local version of the new branch to the developer’s computer.

Data Logging Pertinent internal signals from the FCS are 4. Make the desired changes on a local version of the logged in the PX4 “ULog” format (Ref. 82) on an SD card us- branch and include detailed comments explaining the up- ® ing the UAV Toolbox “PX4 ULog” Simulink block and cus- dates.

tom uORB messages created for RAVEN-SWFT. The logged 5. Perform testing to ensure that the software changes have internal signals included RC commands, sensor data, feed- the desired effect.

back states, control effector commands in engineering and 6. Commit and push the changes to the remote branch of CAN command units, reference commands, and FCS status the software repository.

information. In addition to the internal FCS signals, sensor 7. Create a merge request that includes a detailed descrip- data, internal PX4 INS and external INS states, and PX4 sta- tion of the software changes.

tus messages are stored in the ULog file by specifying these 8. Peer review of the software changes is conducted by at desired topics in the standard PX4 logger topics file.

least one subject matter expert. The assessment includes reviewing the changes relative to the main version of GCS Data Following the approach described in Ref. 51, sig- the repository (e.g., using the visdiff command in ® nals from the FCS are sent to a research GCS via a teleme- MATLAB ), as well as performing software unit tests try radio. In addition to a standard QGroundControl ground and regression tests.

station, the custom research ground control station built in 9. If the software changes are deemed to be successful, the ® Simulink allows for more tailored displays for the particu- software reviewer adds a detailed comment to the GitLab lar aircraft, as well as displays allowing improved monitor- merge request describing the review steps and outcomes.

ing of the flight controller performance and research maneu- 10. The software changes are merged into the main version vers. The research GCS telemetry signals include the vehicle of the GitLab repository by the software reviewer.

This set of procedures upholds a rigorous software vetting flight safety modes that execute an automated response.

process ensuring safety of flight, while also allowing for flexi- bility and rapid development of RAVEN-SWFT research soft- FLIGHT-TEST APPROACH ware.

This section describes the flight-test buildup approach pur- From the RAVEN-SWFT flight authorization viewpoint, it sued for RAVEN-SWFT from the FCS software development is most important to ensure the safety of test personnel and perspective. After simulation development and testing, the ve- bystanders, protect infrastructure outside of the vehicle, and hicle underwent extensive 3DOF testing, followed by tethered keep the vehicle on the flight-test range. Therefore, the most and free flight testing. At the time of writing this paper, the important aspects of the FCS software development, integra- RAVEN-SWFT is actively undergoing the initial phases of its tion, and testing is to ensure that these safety requirements flight testing.

are met. The vehicle’s integrity is still important and every effort is made to avoid harm to the vehicle; however, this is 3DOF Wind-Tunnel Testing considered a secondary priority for the RAVEN-SWFT.

A multilayered flight termination system (FTS) is employed 3DOF free motion wind-tunnel testing was discussed earlier to ensure that the personnel, infrastructure, and range contain- in the paper in the context of aero-propulsive damping charac- ment safety objectives are met. As the first safety mechanism, terization for the RAVEN-SWFT vehicle. Although, dynamic a flight termination switch was implemented on the RC trans- derivative identification to improve the RAVEN-SWFT flight mitter that allows the pilot to enable the FTS flight mode at dynamics model was an important aspect of the 3DOF wind- any time. To prevent accidental tripping of this switch, a lock- tunnel testing, an equally valuable part of the test was focused ing switch was installed on the RC transmitter that requires the on evaluation and refinement of the RAVEN-SWFT FCS soft- pilot to lift and then adjust the switch position to initiate flight ware.

termination. A second safety mechanism is a geofence that Flight testing of new experimental FCS software must pro- enables flight termination if horizontal or vertical boundaries ceed very cautiously because of the elevated risk posed to the are exceeded. The termination geofence acts as an “electronic vehicle. Therefore, substantial effort is invested in perform- tether” from the software assurance and flight authorization ing flight simulations and hardware-in-the-loop bench testing perspective. For both flight termination actions, the vehicle re- before flight testing new FCS software. However, because of sponse is to disable proprotor motor power and set the control differences in the simulation and hardware-in-the-loop envi- surfaces to pro-spin settings. This FTS logic resides within ronment compared to flight, these software verification efforts the unmodified portion of the PX4 firmware and is indepen- can fail to identify possible issues that arise when operating dent of the custom RAVEN-SWFT FCS software that replaces onboard a vehicle in a flight-test environment, which can lead the internal PX4 flight control laws. Nonetheless, to ensure to expensive, unforeseen project delays. 3DOF wind-tunnel the correct response of this critical functionality, termination testing, alternatively, allows the vehicle to be flown in multi- initiation is verified on the ground at the start of each flight ple rotational degrees of freedom in a flight-like environment, day and after re-compiling the custom flight software.

while posing little risk to the vehicle, similar to bench testing.

Additional RAVEN-SWFT flight safety logic includes a lost- The 3DOF wind-tunnel test capability allows the FCS soft- link response and a second geofence inside of the flight ter- ware to be efficiently and rigorously tested, while also per- mination geofence. Currently, in both scenarios, the pro- mitting rapid implementation and testing of software changes grammed response is to automatically return to a hover state based on observations and data from testing. FCS software and land in place. Contrary to the flight termination func- verification constituted a major aspect of the RAVEN-SWFT tionality, the return-to-hover (RTH) capability resides within 3DOF wind-tunnel testing, which accelerated FCS software the custom RAVEN-SWFT FCS software. For the purpose development and served as a substantial risk-reduction effort of flight-test range containment, the RTH response is strongly for subsequent flight testing.

preferred over the FTS response. Similar to the FTS, the lost- The 3DOF wind-tunnel apparatus introduced previously (see link sequence initiation is verified on the ground at the start of Fig. 3), allows for free simultaneous roll, pitch, and yaw mo- each flight day and after re-compiling the custom flight soft- tion where the active control effectors are used to control air- ware. Furthermore, the RTH response was exercised in teth- craft rotational motion. Any of the rotation axes can also ered flights prior to conducting free flight testing.

be fixed for 1DOF or 2DOF testing. The RAVEN-SWFT As a final safety feature, the pilot has the ability in-flight powered-airframe 3DOF wind-tunnel test started by evaluat- to switch to an emergency “Glider Mode” by moving both ing the control laws in a single axis with the other control axes arm switches to their center positions. The vehicle will then disabled and their rotational degrees of freedom locked. Test- switch to direct “stick-to-surface” manual control. Here, the ing then proceeded to include two and three free degrees of lateral, longitudinal, and directional stick positions commands freedom. Ultimately, the custom FCS was able to be rigor- aileron, stabilator, and rudder deflection directly, respectively, ously tested and refined on the 3DOF test apparatus through with the motors disabled. This mode gives the pilot the ability the transition envelope using similar hardware and flight con- to quickly bring the vehicle down to the ground while also re- trol software planned to be used in flight testing. A few of the taining influence over the vehicle’s trajectory, unlike the other minor software differences include: 1. only the inner-loop attitude-stabilization control laws were able to be tested, 2. the pilot stick and FCS reference commands were the roll, pitch, and yaw attitude (the inner-loop 3DOF direc- tional control law used yaw attitude as the feedback vari- able, whereas the flight-test directional control law uses yaw rate as the feedback variable), 3. the transition command was based on the wind-tunnel dynamic pressure setting, and 4. the operational logic was simpler because the vehicle was constrained, did not require takeoff/landing capabil- ities, and the FCS was generally engaged once the wind- tunnel dynamic pressure for the desired test condition Figure 14: RAVEN-SWFT hover 3DOF test setup.

was achieved.

Nonetheless, most of the FCS software was able to be exer- conducted at Vertiport #2 at the NASA LaRC City Environ- cised and refined with minimal risk posed to the vehicle in this ment Range Testing for Autonomous Integrated Navigation important intermediate step prior to conducting flight testing.

(CERTAIN) flight-test range. The tether was suspended above Sufficient inner-loop flight control performance was demon- the vertipad from a large articulating boom lift, as shown in strated in transition throughout the wind-tunnel test envelope, Fig. 15.

in addition to hardware integration verification and vehicle component stress testing, building confidence for flight test- ing.

3DOF Hover Testing After the 3DOF wind-tunnel testing was completed, focus shifted to completing the FCS software updates necessary for flight testing, which included the outer-loop control laws, takeoff/landing logic, and flight safety logic. Then, efforts shifted to verifying the flight-test software updates again on the same 3DOF test apparatus that was used in wind-tunnel tests, but this time in a large, open flight laboratory set- ting using the full flight-test software in FCS Mode 0. The 3DOF apparatus and RAVEN-SWFT aircraft were secured to a heavy metal table, as shown in Fig. 14, and testing was conducted primarily indoors at the Autonomy Lab for Intel- ligent Flight Technology (ALIFT) facility at NASA LaRC.

This setup reduced hover proprotor wake recirculation ef- fects compared to the wind-tunnel test section, allowing for improved hover flight control law evaluation (note that this adverse wind-tunnel proprotor wake recirculation effect only occurs in hover, with the tunnel airflow off—the proprotor wake is blown downstream with the tunnel airflow on, which avoids adverse wind-tunnel proprotor wake recirculation ef- fects). In addition to testing the Mode 0 flight control algo- rithm in hover, the safety features (flight termination response, geofence response, etc.), flight operations logic, and flight-test Figure 15: RAVEN-SWFT tethered flight-test setup.

procedure rehearsals were able to be exercised and refined.

The RAVEN-SWFT tethered flight testing was used to verify Tethered Flight Testing the utility of each flight control mode in hover, the flight-test After completing the 3DOF testing, the vehicle was recon- PTI injection capabilities, and the takeoff/landing logic prior figured for flight-test operations and a tether attachment was to conducting free flight testing without the tether. A photo installed in the center of the wing/fuselage. The tether attach- of the RAVEN-SWFT vehicle in flight on the tether is shown ment was designed to run through the center of gravity of the in Fig. 16. Although the risks to the vehicle in tethered flight vehicle to avoid imparting adverse moments during testing. testing are higher compared to 3DOF testing, tether testing Additionally, a small counterweight was installed on the be- offers the ability to more thoroughly test the full capabilities lay to remove slack from the line. Tethered hover testing was of the hover flight-test FCS algorithm. Also, the vehicle can often be spared by the tether operator in case of unexpected Transition Flight Testing anomalies that could cause substantial damage to the vehicle The next step in the RAVEN-SWFT flight-test campaign is to during free flight testing. Hence, tether testing offered a help- gradually expand testing throughout the transition envelope.

ful opportunity to provide an incremental risk-reduction step At the time of submitting this paper, the RAVEN-SWFT tran- between 3DOF and free flight testing.

sition envelope expansion flight-test campaign has just begun and initial low-speed flights have been completed.

At several conditions throughout the RAVEN-SWFT transi- tion envelope, the flight control performance is planned to be evaluated using piloted and automated doublet maneu- vers. Additionally, flight data will be collected using multi- sine maneuvers to validate and, if necessary, adjust the base- line RAVEN-SWFT transition model. If improved flight con- trol performance is desired at a particular flight condition, the flight control design parameters are planned to be updated us- ing the flight dynamics simulation with aero-propulsive model updates identified from flight testing.

CONCLUDING REMARKS Figure 16: Tethered RAVEN-SWFT hover flight testing.

There are numerous eVTOL aircraft research areas to address prior to bringing these novel vehicles into mainstream opera- tion. The RAVEN-SWFT tiltrotor vehicle has been developed Hover Flight Testing at NASA LaRC to enable advanced eVTOL aircraft model- ing and flight controls research. An attribute of the RAVEN After verifying the hover flight control software in tethered project that distinguishes it from similar pursuits in industry flight testing, the vehicle was detached from the tether and is the objective to openly publish methods and data to facili- proceeded to hover free flight testing. The hover testing was tate collective research advancement helping to enable future composed of piloted assessment of the vehicle response, pi- AAM transportation methods.

loted doublets, automated square wave doublets, and auto- mated multisine maneuvers to assess the performance of mul- This paper provided an overview of the dynamic model- tiple RAVEN-SWFT flight control modes and to collect sys- ing and flight control software development activities for the tem identification flight data enabling validation of the base- RAVEN-SWFT aircraft. Initially, wind-tunnel testing was line RAVEN-SWFT aero-propulsive model. The RAVEN- conducted to develop a high-fidelity flight dynamics simula- SWFT flight-test research started immediately, including en- tion for the aircraft enabling model-based flight control sys- abling the 28 component multisine PTI on the very first free tem design. A robust, well-performing, full-envelope flight flight of the aircraft. A photo of the RAVEN-SWFT perform- control algorithm was designed to enable RAVEN-SWFT test- ing free flight testing is shown in Fig. 17. Note that the metal ing and validation of flight controls research. The flight con- tether connection apparatus was still connected in this photo trol algorithm included an extensive programmed test input (without the tether rope connected) to allow for the flexibility injection capability used to assess the flight control system of switching between tethered and free flight testing, if de- performance and enable efficient system identification flight sired. The tether connection apparatus will be removed after testing. After integrating the flight control software onboard completing hover and low-speed transition testing. the RAVEN-SWFT vehicle, an incremental flight-test build- up approach was employed, including 3DOF free motion wind-tunnel testing and tethered hover flight testing. At the time of submission of this paper, the RAVEN-SWFT has com- pleted its hover test campaign and has just begun transition envelope expansion flight testing. The RAVEN-SWFT flight testing serves as a means of validating modeling and flight controls research that has previously only been demonstrated in simulation and wind-tunnel testing. The RAVEN-SWFT modeling and flight control software development approach has been successful and similar methods are recommended for future subscale aircraft development programs.

ACKNOWLEDGMENTS Figure 17: Initial RAVEN-SWFT hover flight testing.

This research was funded by the NASA Aeronautics Re- search Mission Directorate (ARMD) Transformational Tools and Technologies (TTT) project and Flight Demonstrations 8. Murphy, P. C., Buning, P. G., and Simmons, B. M., and Capabilities (FDC) project. RAVEN-SWFT vehicle de- “Rapid Aero Modeling for Urban Air Mobility Aircraft velopment and testing support was provided by Gregory How- in Computational Experiments,” AIAA SciTech 2021 land, Brayden Chamberlain, Jody Miller, David North, Justin Forum, January 2021. DOI: 10.2514/6.2021-1002.

Lisee, Neil Coffey, and Collin Duke. Additional wind-tunnel 9. Simmons, B. M., Buning, P. G., and Murphy, P. C., test support was provided by Ronald Busan, Stephen Farrell, “Full-Envelope Aero-Propulsive Model Identification Earl Harris, Richard Thorpe, Clinton Duncan, Wes O’Neal, for Lift+Cruise Aircraft Using Computational Experi- Lee Pollard, Stephen Riddick, and Vincent Spada. The 3DOF ments,” AIAA AVIATION 2021 Forum, August 2021.

test apparatus used in this work was designed and built by DOI: 10.2514/6.2021-3170.

Ronald Busan with integration support from personnel in the Flight Dynamics Branch at NASA Langley Research Center.

10. Busan, R. C., Rothhaar, P. M., Croom, M. A., Mur- Other members of the RAVEN project who contributed to phy, P. C., Grafton, S. B., and O’Neal, A. W., “En- the RAVEN-SWFT vehicle development, flight control sys- abling Advanced Wind-Tunnel Research Methods Us- tem development, wind-tunnel testing, and flight testing in- ing the NASA Langley 12-Foot Low Speed Tunnel,” clude: Michael Acheson, Jacob Cook, Jacob Schaefer, Tenavi 14th AIAA Aviation Technology, Integration, and Op- Nakamura-Zimmerer, Siena Whiteside, and Jason Welstead.

erations Conference, June 2014. DOI: 10.2514/6.2014- Discussions with Julia Brault, Ronal George, Ankur Bose, 3000.

Arun Mathamkode, Devin Cote, and Cameron Bosley from The MathWorks, Inc. under Nonreimbursable Space Act 11. Murphy, P. C., and Landman, D., “Experiment Design Agreement SAA1-38060 helped to refine the RAVEN-SWFT for Complex VTOL Aircraft with Distributed Propul- control system implementation approach. The contributions sion and Tilt Wing,” AIAA Atmospheric Flight Mechan- from all team members are gratefully acknowledged and ap- ics Conference, January 2015. DOI: 10.2514/6.2015- preciated.

0017.

REFERENCES 12. Busan, R. C., Murphy, P. C., Hatke, D. B., and Simmons, B. M., “Wind Tunnel Testing Techniques for a Tan- 1. Johnson, W., Silva, C., and Solis, E., “Concept Vehicles dem Tilt-Wing, Distributed Electric Propulsion VTOL for VTOL Air Taxi Operations,” AHS Technical Confer- Aircraft,” AIAA SciTech 2021 Forum, January 2021.

ence on Aeromechanics Design for Transformative Ver- DOI: 10.2514/6.2021-1189.

tical Flight, January 2018.

2. Silva, C., Johnson, W., Patterson, M. D., and Antcliff, 13. Simmons, B. M., and Murphy, P. C., “Aero-Propulsive K. R., “VTOL Urban Air Mobility Concept Vehicles Modeling for Tilt-Wing, Distributed Propulsion Air- for Technology Development,” 2018 Aviation Technol- craft Using Wind Tunnel Data,” Journal of Air- ogy, Integration, and Operations Conference, June 2018.

craft , Vol. 59, (5), September 2022, pp. 1162–1178.

DOI: 10.2514/6.2018-3847.

DOI: 10.2514/1.C036351.

3. Saeed, A. S., Younes, A. B., Cai, C., and Cai, G., “A 14. Simmons, B. M., Morelli, E. A., Busan, R. C., Hatke, survey of hybrid Unmanned Aerial Vehicles,” Progress D. B., and O’Neal, A. W., “Aero-Propulsive Modeling in Aerospace Sciences , Vol. 98, April 2018, pp. 91–105.

for eVTOL Aircraft Using Wind Tunnel Testing with DOI: 10.1016/j.paerosci.2018.03.007.

Multisine Inputs,” AIAA AVIATION 2022 Forum, June 4. Kim, H. D., Perry, A. T., and Ansell, P. J., “A Review of 2022. DOI: 10.2514/6.2022-3603.

Distributed Electric Propulsion Concepts for Air Vehicle 15. Simmons, B. M., and Busan, R. C., “Statisti- Technology,” 2018 AIAA/IEEE Electric Aircraft Tech- cal Wind-Tunnel Experimentation Advancements for nologies Symposium, July 2018. DOI: 10.2514/6.2018- eVTOL Aircraft Aero-Propulsive Model Develop- 4998.

ment,” AIAA SciTech 2024 Forum, January 2024.

5. Johnson, W., and Silva, C., “NASA Concept Vehicles DOI: 10.2514/6.2024-2482.

and the Engineering of Advanced Air Mobility Air- craft,” The Aeronautical Journal , Vol. 126, (1295), Jan- 16. Simmons, B. M., Ackerman, K. A., and Asper, G. D., uary 2022, pp. 59–91. DOI: 10.1017/aer.2021.92.

“Aero-Propulsive Damping Characterization for eV- TOL Aircraft Using Free Motion Wind-Tunnel Test- 6. “eVTOL Aircraft Directory,” Electric VTOL News™ , ing,” AIAA SciTech 2025 Forum, January 2025.

https://evtol.news/aircraft, Accessed 02 March 2025.

DOI: 10.2514/6.2025-0006.

7. Hirschberg, M., “Press Release: Vertical Flight So- ciety Electric VTOL Directory Hits 1,100 Designs,” 17. Simmons, B. M., “System Identification Approach for https://evtol.news/news/press-release-vertical-flight- eVTOL Aircraft Demonstrated Using Simulated Flight society-electric-vtol-directory-hits-1100-designs, Data,” Journal of Aircraft , Vol. 60, (4), July 2023, Accessed 22 March 2025. pp. 1078–1093. DOI: 10.2514/1.C036896.

18. Simmons, B. M., Rapsomanikis, A., Jacobellis, G., Tilt-Wing Aircraft Testing, Modeling, Simulation, Con- Ofodile, N., Hamilton, T., and Ma, M., “Flight-Test Sys- trol, and Flight Test Development,” 14th AIAA Avi- tem Identification Methodology and Hover Results for a ation Technology, Integration, and Operations Confer- Vectored-Thrust eVTOL Aircraft,” Vertical Flight Soci- ence, June 2014. DOI: 10.2514/6.2014-2999.

ety’s 81st Annual Forum & Technology Display, May 29. McSwain, R. G., Geuther, S. C., Howland, G., Patterson, 2025, To be published.

M. D., Whiteside, S. K., and North, D. D., “An Experi- 19. Cook, J., and Gregory, I., “A Robust Uniform Con- mental Approach to a Rapid Propulsion and Aeronautics trol Approach for VTOL Aircraft,” VFS Autonomous Concepts Testbed,” NASA TM–2020-220437, January VTOL Technical Meeting and Electric VTOL Sympo- 2020.

sium, January 2021.

30. North, D. D., Busan, R. C., and Howland, G., 20. Acheson, M. J., Gregory, I. M., and Cook, J., “Examina- “Design and Fabrication of the Langley Aero- tion of Unified Control Incorporating Generalized Con- drome No. 8 Distributed Electric Propulsion VTOL trol Allocation,” AIAA SciTech 2021 Forum, January Testbed,” AIAA SciTech 2021 Forum, January 2021.

2021. DOI: 10.2514/6.2021-0999.

DOI: 10.2514/6.2021-1188.

21. Milz, D., May, M. S., and Looye, G., “Dynamic 31. Cooper, J. R., Ackerman, K. A., Rothhaar, P. M., and Inversion-Based Control Concept for Transformational Gregory, I. M., “Autonomous Path-Following for a Tilt- Tilt-Wing eVTOLs,” AIAA SciTech 2024 Forum, Jan- Wing, Distributed Electric Propulsion, Vertical Take-Off uary 2024. DOI: 10.2514/6.2024-1290.

and Landing Unmanned Aerial System in Hover Mode,” NASA TM–2018–220109, November 2018.

22. Comer, A., and Chakraborty, I., “Total Energy-Based Control Design and Optimization for Lift-Plus-Cruise 32. McSwain, R. G., Glaab, L. J., and Theodore, C. R., Aircraft,” Journal of Guidance, Control, and Dy- “Greased Lightning (GL-10) Performance Flight Re- namics , Vol. 47, (7), July 2024, pp. 1414–1436.

search – Flight Data Report,” NASA TM–2017–219794, DOI: 10.2514/1.G007605.

November 2017.

23. Comer, A. M., and Chakraborty, I., “Full Envelope 33. Fredericks, W. J., McSwain, R. G., Beaton, B. F., Klass- Flight Control System Design and Optimization for man, D. W., and Theodore, C. R., “Greased Lightning a Tilt-Wing Aircraft,” Journal of the American He- (GL-10) Flight Testing Campaign,” NASA TM–2017– licopter Society , Vol. 69, (3), July 2024, pp. 1–18.

219643, July 2017.

DOI: 10.4050/JAHS.69.032003.

34. Geuther, S. C., and Fei, X., “LA-8 Computa- 24. Comer, A. M., “Optimizing Explicit Model-Following tional Analysis and Validation Studies Using Flight- Trajectory Control Laws for a Vectored Thrust Stream,” AIAA SciTech 2021 Forum, January 2021.

Conguration,” Journal of the American Helicopter DOI: 10.2514/6.2021-1191.

Society , Vol. 70, (1), January 2025, pp. 1–20.

DOI: 10.4050/JAHS.70.012007.

35. Ahuja, V., and Hartfield, R. J., “Reduced-Order Aerodynamics with the Method of Integrated Circu- 25. Geuther, S. C., Simmons, B. M., and Ackerman, K. A., lation,” AIAA SciTech 2022 Forum, January 2022.

“Overview of the Subscale RAVEN Flight Controls and DOI: 10.2514/6.2022-0005.

Modeling Testbed,” Vertical Flight Society’s 80th An- nual Forum & Technology Display, May 2024.

36. DiMaggio, G. A., Simmons, B. M., Geuther, S. C., Hartfield, R. J., and Ahuja, V., “Transition Aero- 26. German, B. J., Jha, A., Whiteside, S. K. S., and Propulsive Analysis of a Tilt-Wing eVTOL Aircraft Us- Welstead, J. R., “Overview of the Research Air- ing a Surface-Vorticity Solver,” AIAA SciTech 2025 Fo- craft for eVTOL Enabling techNologies (RAVEN) Ac- rum, January 2025. DOI: 10.2514/6.2025-0654.

tivity,” AIAA AVIATION 2023 Forum, June 2023.

DOI: 10.2514/6.2023-3924.

37. Geuther, S. C., North, D. D., and Busan, R. C., “In- vestigation of a Tandem Tilt-Wing VTOL Aircraft in 27. “NASA Langley 12-Foot Low-Speed Tunnel,” the NASA Langley 12-Foot Low-Speed Tunnel,” NASA https://researchdirectorate.larc.nasa.gov/12-foot- TM–2020–5003178, June 2020.

low-speed-tunnel-12-ft-lst/, Accessed 17 September 2024.

38. Simmons, B. M., and Hatke, D. B., “Investigation of 28. Rothhaar, P. M., Murphy, P. C., Bacon, B. J., Gre- High Incidence Angle Propeller Aerodynamics for Sub- gory, I. M., Grauer, J. A., Busan, R. C., and Croom, scale eVTOL Aircraft,” NASA TM–20210014010, May M. A., “NASA Langley Distributed Propulsion VTOL 2021.

39. Simmons, B. M., “System Identification for Pro- 52. Fritsch, F. N., and Carlson, R. E., “Monotone Piece- pellers at High Incidence Angles,” Journal of Air- wise Cubic Interpolation,” SIAM Journal on Nu- craft , Vol. 58, (6), November 2021, pp. 1336–1350. merical Analysis , Vol. 17, (2), 1980, pp. 238–246.

DOI: 10.2514/1.C036329. DOI: 10.1137/0717021.

53. Moler, C. B., Numerical Computing with 40. Stratton, M., and Landman, D., “Wind Tunnel Test and MATLAB , Society for Industrial and Applied Empirical Modeling of Tilt-Rotor Performance for eV- Mathematics, Philadelphia, PA, January 2004.

TOL Applications,” AIAA SciTech 2021 Forum, Jan- DOI: 10.1137/1.9780898717952.

uary 2021. DOI: 10.2514/6.2021-0834.

54. Morelli, E. A., and Klein, V., Aircraft System Iden- 41. Simmons, B. M., “Evaluation of Response Surface Ex- tification: Theory and Practice , Sunflyte Enterprises, periment Designs for Distributed Propulsion Aircraft Williamsburg, VA, second edition, 2016.

Aero-Propulsive Modeling,” AIAA SciTech 2023 Fo- rum, January 2023. DOI: 10.2514/6.2023-2251. 55. Etkin, B., and Reid, L. D., Dynamics of Flight: Stability and Control , John Wiley & Sons, Inc, New York, NY, 42. Simmons, B. M., “Efficient Variable-Pitch Propeller third edition, 1996.

Aerodynamic Model Development for Vectored-Thrust 56. Stevens, B. L., Lewis, F. L., and Johnson, E. N., Aircraft eVTOL Aircraft,” AIAA AVIATION 2022 Forum, June Control and Simulation: Dynamics, Controls Design, 2022. DOI: 10.2514/6.2022-3817.

and Autonomous Systems , John Wiley & Sons, Hobo- ken, NJ, third edition, 2015.

43. Simmons, B. M., Geuther, S., and Ahuja, V., “Valida- tion of a Mid-Fidelity Approach for Aircraft Stability 57. Pei, J., and Roithmayr, C., “Equations of Motion for and Control Characterization,” AIAA AVIATION 2023 a Generic Multibody Tilt-rotor Aircraft,” AIAA AVIA- Forum, June 2023. DOI: 10.2514/6.2023-4076.

TION 2022 Forum, June 2022. DOI: 10.2514/6.2022- 3511.

44. Montgomery, D. C., Design And Analysis of Experi- ments , John Wiley & Sons, Inc., Hoboken, NJ, 8th edi- 58. “U.S. Standard Atmosphere,” NOAA–S/T 76–1562, Oc- tion, 2013.

tober 1976.

45. Myers, R. H., Montgomery, D. C., and Anderson- 59. Johnson, D. L., “Terrestrial Environment (Climatic) Cri- Cook, C. M., Response Surface Methodology: Process teria Guidelines for Use in Aerospace Vehicle Develop- and Product Optimization Using Designed Experiments , ment, 1993 Revision,” NASA TM–4511, August 1993.

John Wiley & Sons, Hoboken, NJ, fourth edition, 2016.

60. “Flying Qualities of Piloted Aircraft,” MIL-STD- 1797A, January 1990.

46. Vicroy, D. D., Loeser, T. D., and Sch¨ utte, A., “Static and Forced-Oscillation Tests of a Generic 61. Chen, C.-T., Linear System Theory and Design , Oxford Unmanned Combat Air Vehicle,” Journal of Air- University Press, fourth edition, 2014.

craft , Vol. 49, (6), November 2012, pp. 1558–1583.

62. Young, P. C., and Willems, J., “An Approach to the Lin- DOI: 10.2514/1.C031501.

ear Multivariable Servomechanism Problem,” Interna- 47. Vicroy, D. D., “A Guide to Forced Oscillation Data Pro- tional Journal of Control , Vol. 15, (5), 1972, pp. 961– cessing and Analysis,” NASA TP–20210023569, June 979.

2023.

63. Berger, T., “Handling Qualities Requirements and Con- trol Design for High-Speed Rotorcraft,” Special Report 48. “Flight Dynamics Simulation of a Generic Transport FCDD-AMV-20-01, February 2020.

Model,” NASA Technology Transfer Program, https: //software.nasa.gov/software/LAR-17625-1, Accessed 64. Enns, D., “Control Allocation Approaches,” Guidance, 17 September 2024.

Navigation, and Control Conference and Exhibit, Au- gust 1998. DOI: 10.2514/6.1998-4109.

49. “MathWorks Documentation,” https://www.mathworks.

com/help/, Accessed 01 March 2025.

65. Acheson, M. J., and Gregory, I. M., “Modi- fied Cascading Generalized Inverse Control Alloca- 50. “X-Plane,” https://www.x-plane.com/, Accessed 22 tion,” AIAA SciTech 2023 Forum, January 2023.

March 2025.

DOI: 10.2514/6.2023-2542.

51. Comer, A. M., Simmons, B. M., and Asper, 66. Berger, T., Tischler, M. B., Hagerott, S. G., Gangsaas, G. D., “Design, Simulation, and Flight Testing of a D., and Saeed, N., “Longitudinal Control Law De- Multi-Purpose VTOL Flight Control System,” NASA sign and Handling Qualities Optimization for a Busi- TM–20250000954, To be published. ness Jet Flight Control System,” AIAA Guidance, Navigation and Control Conference, August 2012. 79. Simmons, B. M., Gresham, J. L., and Woolsey, C. A., DOI: 10.2514/6.2012-4503. “Flight-Test System Identification Techniques and Ap- plications for Small, Low-Cost, Fixed-Wing Aircraft,” 67. Stoll, A., and Bevirt, J., “Development of eVTOL Air- Journal of Aircraft , Vol. 60, (5), September 2023, craft for Urban Air Mobility at Joby Aviation,” Vertical pp. 1503–1521. DOI: 10.2514/1.C037260.

Flight Society’s 78th Annual Forum & Technology Dis- play, May 2022. DOI: 10.4050/F-0078-2022-17528. 80. “QGroundControl,” https://qgroundcontrol.com/, Ac- cessed 23 March 2025.

68. Jategaonkar, R. V., Flight Vehicle System Identifica- 81. DroneCAN, “DroneCAN Documentation,” https:// tion: A Time-Domain Methodology , American Institute dronecan.github.io/, Accessed 18 April 2024.

of Aeronautics and Astronautics, Reston, VA, second edition, 2015. DOI: 10.2514/4.102790.

82. “ULog File Format,” https://docs.px4.io/main/en/dev log/ulog file format.html, Accessed 12 March 2025.

69. Tischler, M. B., and Remple, R. K., Aircraft and Rotor- craft System Identification: Engineering Methods With 83. “GitLab,” https://about.gitlab.com/, Accessed 09 March Flight Test Examples , American Institute of Aeronau- 2025.

tics and Astronautics, Reston, VA, second edition, 2012.

DOI: 10.2514/4.868207.

70. “System IDentification Programs for AirCraft (SID- PAC),” NASA Technology Transfer Program , https: //software.nasa.gov/software/LAR-16100-1, Accessed 08 November 2024.

71. Morelli, E. A., “Multiple Input Design for Real-Time Parameter Estimation in the Frequency Domain,” 13th IFAC Conference on System Identification, August 2003. DOI: 10.1016/S1474-6670(17)34833-4.

72. Morelli, E. A., “Flight-Test Experiment Design for Characterizing Stability and Control of Hypersonic Vehicles,” Journal of Guidance, Control, and Dy- namics , Vol. 32, (3), May 2009, pp. 949–959.

DOI: 10.2514/1.37092.

73. Morelli, E. A., “Practical Aspects of Multiple- Input Design for Aircraft System Identification Flight Tests,” AIAA AVIATION 2021 Forum, August 2021.

DOI: 10.2514/6.2021-2795.

74. Grauer, J. A., and Boucher, M., “System Identification of Flexible Aircraft: Lessons Learned from the X-56A Phase 1 Flight Tests,” AIAA SciTech 2020 Forum, Jan- uary 2020. DOI: 10.2514/6.2020-1017.

75. Asper, G. D., and Simmons, B. M., “Rapid Flight Control Law Deployment and Testing Framework for Subscale VTOL Aircraft,” NASA TM–20220011570, September 2022.

76. Asper, G. D., Simmons, B. M., Axten, R. M., Acker- man, K. A., and Corrigan, P. E., “Inexpensive Multirotor Platform for Advanced Controls Testing (IMPACT): De- velopment, Integration, and Experimentation,” NASA TM–20240000223, March 2024.

77. “PX4 Autopilot User Guide,” https://docs.px4.io/main/ en/, Accessed 11 March 2025.

78. “uORB Messaging,” https://docs.px4.io/main/en/ middleware/uorb.html, Accessed 18 April 2024.