UAV Research at Georgia Tech

Transcription

1 UAV Research at Georgia Tech Eric N. Johnson Lockheed Martin Assistant Professor of Avionics Integration, Georgia Tech School of Aerospace Engineering Presentation at TU Delft June 3, 2002 June 2002 ENJ - Georgia Tech 1

2 Previous Work Outline MIT and Draper Laboratory Ph.D. Thesis Work: Advanced Control for the X-33X Current Research Adaptive Guidance and Control for Hypersonic Vehicles Aggressive Maneuvering for UAVs DARPA Software Enabled Control, and the GTMax UAV Aerial Robotics Competition June 2002 ENJ - Georgia Tech 2

3 Draper Small Autonomous Air Vehicle (DSAAV) in 1996 IMU SD MotionPak Receiver/Servo Interface Battery Compass 6 ft Rotor Sonar Altimeter D-GPS NovAtel RT-20 Camera/Tx RF Modem 32cc Engine Modified TSK BlackStar Total Weight 23 Pounds, 10 kg 486 Computer Power Distribution June 2002 ENJ - Georgia Tech 3

4 DSAAV at the 1996 Aerial Robotics Competition Organized by the Association for Unmanned Vehicle Systems, International (AUVSI) Epcot Center, Orlando, Florida D-GPS Reference Contest Area, 60x120 ft Vision Processor GCS Start Box Helicopter Ground Coverage l Ae ria R ob ot ic s Emergency Termination Safety Pilot st on o B Tea m June 2002 ENJ - Georgia Tech 4

5 Contest Flight #5 June 2002 ENJ - Georgia Tech 5

6 Limited Authority Adaptive Flight Control Research Project Sponsored by NASA MSFC Thesis Advisor: Anthony J. Calise, Georgia Tech Exploring Flight Control Technologies Applicable to X-33 & Future Reusable Launch Vehicles (RLV) Reduce Analysis Required per Mission Increase Tolerance to Failures and Environment June 2002 ENJ - Georgia Tech 6

7 Neural-Network Adaptive Flight Control Command Pseudo-Control Plant Inputs (Actual Controls) Reference Model ν rm + ν Approximate Dynamic Inversion δ Plant + - Tracking Error ν pd PD PD Control ν ad Neural Network June 2002 ENJ - Georgia Tech 7

8 Single Hidden Layer Neural Network Feedforward Neural Networks with a Single Hidden Layer are Universal Approximators. x 1 V () σ () σ W y 1 The Sigmoidal Activation Function has Internal Activation Potential a. σ ( z) az 1 = 1 + e x 2 x N1 σ () N 1 N 3 o o () σ N 2 o y 2 y N3 In matrix form: ν ad = y = T W σ ( T V x) June 2002 ENJ - Georgia Tech 8

9 Neural Network Adaptation Error Dynamics: ν = ν x e crm = ν rm + ν pd ν ad ( ν ) = Aerm + b ad e rm x = x (A is Hurwitz) rm rm x x Define: ζ = Pb e T rm A T P + PA = Q, Q > 0 Adaptation Law: WD = Γ VD = [( T ) ] W σ σ' V x ζ + κ ζ W [ T xζw σ' + κ ζ V] Γ V σ' () z = σ z () z (Diagonal Matrix) June 2002 ENJ - Georgia Tech 9

10 Capability is Limited Issues Saturation (Including Axis Priority), Rate Limits Not Feedback Linearizable Sign of Control Effectiveness Becomes Zero Discrete Control (e.g., RCS Thrusters) Need to Make a Flight Certification Case Show Adaptation Extremely Unlikely to Cause Loss of Vehicle Assumptions for Stability Need to be Extremely Mild Require Recovery from Temporary Faulty Adaptation June 2002 ENJ - Georgia Tech 10

11 NN Adaptive Control with Pseudo-Control Hedging (PCH) Command ν hedge Estimate Hedge x Reference Model ν rm + ν Dynamic Inversion δ cmd Actuator δ Plant + - ν pd PD PD Control ν ad Neural Network Tracking Error June 2002 ENJ - Georgia Tech 11

12 Implications Shelter Adaptive Element from the Adverse Effects of Plant Input Characteristics: Linear Dynamics, Latency, Saturation, Rate Saturation, etc. Achievable Adaptation Performance is Increased Dramatically Adaptation is Correct During Saturation Adaptive Element Can Recover from Faulty Adaptation Enables Correct Adaptation When Not in Control of Plant June 2002 ENJ - Georgia Tech 12

13 Ascent Phase X-33 Flight Control Sponsored by NASA MSFC Linear Aerospike Roll/Pitch/Yaw Aerodynamic Controls: Body Flaps Elevons Rudders Transition and Entry Reaction Control System (RCS) Aerodynamic Controls Aero Surfaces (8) RCS (8) Aerospike Throttles (4) June 2002 ENJ - Georgia Tech 13

14 Nominal Ascent Phase Results Preliminary Results, Ascent Flight Control 3-Axis Attitude System Performance Improved Over Existing Design Attitude Error is Lower Hinge Moments Look Good Nothing is Scheduled! Baseline attitude error (deg) NN roll pitch yaw time (sec) roll pitch yaw time (sec) June 2002 ENJ - Georgia Tech 14

15 Ascent Phase Multiple Actuator Failures Half of Aero Surfaces Fail Hard-Over at 60 sec (All Right-Hand Surfaces Give Uncommanded Left Turn) Occurs Near Max Q (60 Seconds) Baseline NN Failure 120 roll pitch yaw time (sec) roll pitch yaw attitude error (deg) time (sec) June 2002 ENJ - Georgia Tech 15

16 Ascent Phase Multiple Actuator Failures NN Controller Saturates on All Three Axes Vehicle Rolls Three Times Full Recovery Once Dynamic Pressure Drops surface deflection (deg) flapr flapl elevoninr elevoninl elevonoutr elevonoutl rudderr rudderl time (sec) Effectors June 2002 ENJ - Georgia Tech 16

17 Adaptation is Correct During Saturation No Knowledge of Failure Used (Not Even in the Hedge!) Ascent Phase Multiple Actuator Failures NN Controller Roll Axis Pseudo-Control Signals del vad time (sec) June 2002 ENJ - Georgia Tech 17

18 Subsequent Research Involving PCH X-33/RLV Attitude Control Adaptive Tracking and Control (Inner and Outer Loops) for RLV Reconfigurable Flight Control for Civillian Aircraft (Training While Not in Control) Yamaha R-50/RR 50/R-Max JDAM June 2002 ENJ - Georgia Tech 18

19 Georgia Tech UAV Research Facility June 2002 ENJ - Georgia Tech 19

20 Approach to Adaptive Trajectory Following PCH is Used To Modify the Command Trajectory to Create the Feasible Reference Trajectory (And Leave it Alone if Not at Limits) Protect Outer Loop Adaptation From Inner Loop Dynamics Protect Inner Loop Adaptation From Limited Control Authority (As Before) PCH PCH Command Trajectory Outer Outer Loop Loop x, v θ Inner Inner Loop Loop Neural Neural Network June 2002 ENJ - Georgia Tech 20

21 Application to Rotorcraft Maneuvering 80 Network ON 60 Yamaha R-Max Simulation Results: Fly in a Circle While Pirouetting North st time around Network OFF Pentagon Network OFF North Circle Network ON Better Each Time East Vel = 15 ft/s Yaw = 45 o /sec East June 2002 ENJ - Georgia Tech 21

22 Software Enabled Control Sponsored by DARPA Develop software-enabled enabled control methods for complex dynamic systems with application focus on intelligent UAVs Support-the the-development and implement a plug-and and- play, real-time software architectures VTOL UAV hardware-in in-the-loop simulation and flight testing June 2002 ENJ - Georgia Tech 22

23 Bold Stroke Open Systems Architecture Real-time CORBA-based Integration of Distributed, Heterogeneous Components Utilizes Object Request Broker (ORB) Architecture Developed by Washington University and Boeing Application Component Application Component Application Component Services Time Services Non-Volatile Memory Services Scheduling Services Timer Services Event Services Real-Time ORB OS and Hardware Interfaces Naming Services Persistence Services June 2002 ENJ - Georgia Tech 23

24 Component Communication Example PID Neural Net Controller Strategy Distributed objects Plug-and and-play Encapsulation Reconfiguration Controller Interface Open Systems Architecture UAV Interface UAV Interface Simulation Model Vehicle Sensor Interfaces Sensor Models Rigid Body + Rotor Dynamics Actuator Interface Sensor Interfaces Actuator Interface Servo Dynamics Force and Moment Calculations June 2002 ENJ - Georgia Tech 24

25 Recent UAV Platform Integration Work Yamaha R-Max, R 66kg, 3m Rotor Diameter Avionics and Simulation Tools Developed Over the Past Year Hardware-in in-the-loop Simulation and Ground Testing Started in November 2001 Navigation System Ground Tests Completed February 2002 Flights Testing (With Avionics) Began March 2002 June 2002 ENJ - Georgia Tech 25

26 GTMax Hardware Components Flight Computer 266MHz Embedded PC, Ethernet, Flash Drive Sensors Inertial Measurement Unit Differential GPS Magnetometer Sonar and Radar Altimeters Vehicle Telemetry (RPM, Voltage, Pilot Inputs) Data Links 11 Mbps Ethernet Data Link RS-232 Serial Data Link June 2002 ENJ - Georgia Tech 26

27 GTMax Hardware Integration Exchangeable modules: Flight Computer Module GPS Module Data Link Module IMU/Radar Module Unused Module (Growth) Sonar/Magnetometer Assemblies Power Distribution System Each module has self- contained power regulation and EMI shielding Vibration isolated main module rack June 2002 ENJ - Georgia Tech 27

28 Onboard Avionics Architecture GPS Module NovAtel RT-2 GPS Receiver 5V DC/DC 5V HMR-2300 Magnetometer Sonar Altimeter IMU/Radar Module ISIS-IMU Radar Altimeter DC/DC 12V Power Distribution Module Flight Computer Module Serial Extension Board Flight Computer 5V DC/DC 12V Battery 12V Generator Data Link Module Aironet MC4800 Freewave DGR-115 Ethernet Hub Yamaha Attitude Control System RC Receiver YACS IMU Auxiliary Module Auxiliary Computer / Payload 5V DC/DC 12V RS-232 Serial Ethernet DC Power June 2002 ENJ - Georgia Tech 28

29 Baseline Onboard Software Navigation 17 State Extended Kalman Filter Navigation System Vehicle Position Vehicle Velocity Vehicle Attitude Accelerometer Biases Gyro Biases Terrain Height All Attitude Capable 100 Hz Updates Flight Operational Control Adaptive Neural Network Trajectory Following Controller Neural Network 16 Inputs 5 Hidden Layer Neurons 6 Outputs for 6 Degrees of Freedom Can Also Be Configured as a Conventional Inverting Controller Flight Operational June 2002 ENJ - Georgia Tech 29

30 Hardware In the Loop Simulation Capable The Desktop Computer Simulation Utilizes Actual Flight Software Actual Ground Control Station Software Flight Test Verified Dynamic Model of Helicopter Flight Test Verified Model of All Sensors/Actuators Scene Generation Capability Simulation Tools June 2002 ENJ - Georgia Tech 30

31 Software in the Loop (SITL) Test algorithms within the simulation Generate emulated sensor data from an aircraft simulation (including errors) Desktop Computer Sensor Emulation (w/ Error Model) State Vehicle Model Control Actuator Simulation Sensor Raw Data Actuator Raw Data Sensor Drivers Sensor Data Navigation Filter State Estimate Flight Controller Control Actuator Driver Other Systems Trajectory Planner June 2002 ENJ - Georgia Tech 31

32 Hardware in the Loop (HITL) Flight software runs on the onboard computer Onboard computer thinks it is flying the vehicle Desktop Computer Sensor Emulation (w/ Error Model) State Vehicle Model Control Actuator Simulation Sensor Raw Data Actuator Raw Data Sensor Drivers Sensor Data Navigation Filter State Estimate Flight Controller Control Actuator Driver Flight Computer Other Systems Trajectory Planner June 2002 ENJ - Georgia Tech 32

33 GTMax Flight Operations Network Connections Available At Ground Control Station from Hub Multiple Laptops Can Communicate with Onboard Computers Simultaneously Due to Generator, Endurance Limited by Onboard Fuel (> 1 hour) Ground Equipment Can Operate on 115VAC or 12VDC and Has Battery Backup June 2002 ENJ - Georgia Tech 33

34 Flight Testing in McDonough, Georgia June 2002 ENJ - Georgia Tech 34

35 First Tests w/ Baseline Controller Neural Network Adaptive Controller on First Flight Test Day (April 10, 2002) Even With Large Model Errors, System Was Able To Control the Helicopter June 2002 ENJ - Georgia Tech 35

36 Results With Baseline Controller Step Input of Altitude Command: step input of position command, down 170 P os ition Estima te P os ition Comma nd 168 altitude (ft) time ( se c ) June 2002 ENJ - Georgia Tech 36

37 Flight Control Reconfigurations Switched Between Neural Network Adaptive Controller to Much Simpler Conventional Inverting Controller and Back Real Time and Closed Loop Control Reconfiguration P osition Estima te Position Command East (ft) Reconfiguration at time (se c ) June 2002 ENJ - Georgia Tech 37

38 Simulated Main Rotor Actuator Failure With fault tolerant and reconfigurable control system Failure detection and control reconfiguration with RPM control Collective control failure Without fault tolerant and reconfigurable control system June 2002 ENJ - Georgia Tech 38

39 Tail Rotor Failure (in Simulation) With fault tolerant and reconfigurable control system Translatory descent to a clear area Control reconfiguration using main rotor controls Gain altitude using main rotor collective Control reconfiguration for autorotation Autorotation and landing Tail rotor failure Without fault tolerant and reconfigurable control system June 2002 ENJ - Georgia Tech 39

40 International Aerial Robotics In Less Than 15 Minutes: Sign Over Entry 3 km Two Lights Identify Building Building and an Entry Point Found Launch Area Image Receiver (& Other Gound Components) >1m Vehicle or Subvehicle(s) Enter Building Transmit an Image of Point of Interest Inside Building June 2002 ENJ - Georgia Tech 40

41 International Aerial Robotics Competition Unmanned and Autonomous (No Active Human Operators, no Tethers) Some Components Can Remain on the Ground (e.g., Additional Computers, Navigation Aids) Launch and Recovery Need Not Be Autonomous Mission is Divided into Levels Each Teams Gets 60 Minutes To Fly (...Per Year) Rules Change Once a Mission is Completed avdil.gtri.gatech.edu/auvs/ /AUVS/CurrentIARC/2001CollegiateRules.html June 2002 ENJ - Georgia Tech 41

42 Mission Levels Level 1: Follow Prescribed Waypoints for 3km Level 2: Locate Building and Find an Entry Level 3: Enter the Building Can Be a Different Vehicle or Subvehicle That Used Above Can Launch Near Target Structure Level 4: Image Desired Location Within Building and Transmit Complete In < 15 Minutes (Launch to Data Retrieval) Contest is Over Once Somebody Does Level 4 June 2002 ENJ - Georgia Tech 42

43 2001 Airplane: ¼ Scale Cub June 2002 ENJ - Georgia Tech 43

44 Winning Flight in 2001, Level 1 Altitude Waypoint 1 Runway & Ground Station Waypoint 3 Waypoint 4 & Holding Pattern Latitude Under Automatic Flight: Distance Traveled: 3.1 mi / km Time: 3 min 9 sec Average Speed: 58 mph / 93 kph Max Distance from Ground Station: ½ mi / 0.8 km Average Altitude: 397 ft / 121 m Longitude Waypoint 2 Automatic Flight Manual Takeoff/Landing June 2002 ENJ - Georgia Tech 44

45 Level 2 Plans for 2002 Add a Video Camera and Image Processor (Donation from Texas Instruments) Switch GPS to D-GPS D For Level 2 Accuracy (NovAtel) Update Ground Station Software and Develop Image Processing Software Possibly Also Switch to GTMax Level 2+ Design, Building, and Testing for Sub-Vehicle: Drops From Airplane and Enters Building Designs for Operation Inside Building (Levels 3 & 4) June 2002 ENJ - Georgia Tech 45

46 Some Potential Areas of Collaboration VTOL and Fixed-Wing UAV Flight Testing Lessons Learned Simulation Models and Software GTMax as a Research Flight Test Platform Studies at Georgia Tech June 2002 ENJ - Georgia Tech 46

47 UAV Avionics at the 2002 Digital Avionics Systems Conference AIAA/IEEE October 27-31; Irvine, California NEW Applications of Avionics: Uninhabited Air Vehicles (UAV) & Missiles Track: Avionics systems for UAVs, intelligent systems for vehicle autonomy, omy, operation of UAVs in controlled airspace, payloads, missiles, and guided munitions 5 Sessions Paper Acceptance Still Possible for New Track, But Act Fast Contact: Eric N. Johnson , Eric.Johnson@ae ae.gatech.eduedu June 2002 ENJ - Georgia Tech 47