Learning Parameterized Maneuvers for Autonomous Helicopter Flight

1. Learning Parameterized Maneuvers for Autonomous Helicopter Flight Jie Tang, Arjun Singh, Nimbus Goehausen, Pieter Abbeel UC Berkeley

2. Overview Dynamics Model Controller Optimal Control Target Trajectory

3. Problem • Robotics tasks involve complex trajectories • Stall turn • Challenging, nonlinear dynamics

4. Overview Demonstrations Dynamics Model Controller Optimal Control Target Trajectory

5. Learning Target Trajectory From Demonstration Problem: Demonstrations are suboptimal • Use multiple demonstrations • Current state of the art in helicopter aerobatics (Coates, Abbeel, and Ng, ICML 2008) Problem: Demonstrations will be different from desired target trajectory • Our work: learn parameterized maneuver classes Height

6. ExampleData

7. Learning Trajectory Height 50m Hidden • HMM-like generative model • Dynamics model used as HMM transition model • Synthetic observations enforce parameterization • Demos are observations of hidden trajectory • Problem: how do we align observations to hidden trajectory? Demo 1 Demo 2

8. Learning Trajectory Height 50m Hidden • Dynamic Time Warping • Extended Kalman filter / smoother • Repeat Demo 1 Demo 2

9. Smoothed Dynamic Time Warping • Potential outcome of dynamic time warping: • More desirable outcome: • Introduce smoothing penalty • Extra dimension in dynamic program

10. Weighting Demonstrations • Some demonstrations should contribute more to target trajectory than others • Difficult to tune these observation covariances • Learn optimal observation covariances using EM TargetHeight

11. Learned Trajectory TargetHeight

12. Frequency Sweeps and Step Responses Overview Demonstrations Dynamics Model Controller Optimal Control Target Trajectory

13. Learning dynamics • Standard helicopter dynamics model estimated from data • Has relatively large errors in aggressive flight regimes • After learning target trajectory, we obtain aligned demonstrations • Errors in model are consistent for executions of the same maneuver class • Many hidden variables are not modeled explicitly • Airflow, rotor speed, actuator latency • Learn corrections to dynamics model along each target trajectory 2G error

14. Frequency Sweeps and Step Responses Overview Dynamics Model Demonstrations Trajectory-Specific Corrections Standard Dynamics Model + Controller Optimal Control Optimal Control Receding Horizon Differential Dynamic Programming Target Trajectory

15. Experimental Setup Extended Kalman Filter RHDDP controller “Position” 3-axis magnetometer, accelerometer, gyroscope (“Orientation”) Offboard Cameras 1280x960@20Hz Controls @ 20Hz Onboard IMU @333Hz

16. Results: Stall Turn Max speed: 57 mph

17. Results: Loops

18. Results: Tic-Tocs

19. Typical Flight Performance: Stall Turn

20. Quantitative Evaluation • Flight conditions: wind up to 15mph • Similar accuracy is maintained for queries very different from our demonstrations • e.g., can learn 60m stall turns from 40m, 80m demonstrations • Four or five demonstrations sufficient to cover a wide range of stall turns, loops, and tic-tocs • e.g., four stall turns at 20m, 40m, 60m, 80m sufficient to generate any stall turn between 20m and 80m

21. Conclusions • Presented an algorithm for learning parameterized target trajectories and accurate dynamics models from demonstrations • With few demonstrations, can generate a wide variety of novel trajectories • Validated on a variety of parameterized aerobatic helicopter maneuvers

22. Thank you

23. Thank you