Motion Estimation In High Speed OffRoad Maneuvers Haomiao

Motion Estimation In High Speed Off-Road Maneuvers Haomiao Huang Partner: Lyle Chamberlain Mentor: Prof. Richard Murray Sponsor: Northrop Grumman Oct. 16 th 2004 Haomiao Huang

Motivation: The DARPA Grand Challenge GRAND PRIZE $2, 000!! 2/23/2021 Haomiao Huang 2

The “Powerslide” • Sliding is inevitable when driving on dirt roads • Take advantage of it! • No need to brake • Sharper turning radius • Just plain looks cool! www. driftsession. com 2/23/2021 Haomiao Huang 3

Sub-scale Test Platform: Peggy Key Requirements: • Off-road racing chasis • Easily modified • High performance • Robust to rollovers/crashes • Small, for ease of testing Solution: Modify 1/6 th-scale RC racer 2/23/2021 Haomiao Huang 4

The Sensors GPS IMU • GPS provides position/heading at 1 Hz • Inertial Measurement Unit provides gyro and accelerometer measurements at 50 Hz Embedded System 2/23/2021 • Embedded sensors provide wheel velocities, engine RPM, and RC control inputs at 10 Hz Haomiao Huang 5

State Estimation Goal: Determine the state (position, attitude, sideslip? ) of the vehicle at any moment in time. Real-time processing not yet implemented, so post-process data to obtain vehicle state. 1) Obtain GPS points 2) Fit spline to GPS points to get interpolated velocity and positions 3) Integrate gyro data to get true heading 2/23/2021 Haomiao Huang 6

Heading & Sideslip Estimation • Sideslip angle measured by subtracting velocity angle from gyro heading • Gyro drifts, so it must be corrected from GPS heading vector using a Kalman Filter, but updates can only occur when vehicle is not sliding. V Ф Θ • Key difficulty: deciding when the vehicle is/isn’t sliding • Solution: Gyro drift is small, so keep running estimate of sideslip angle and update when the sideslip angle is small. 2/23/2021 Haomiao Huang Ф=Θ-V 7

Field Testing Human-controlled sliding turns conducted at El Mirage Dry Lake Bed with sensor data logged on laptop. Maneuvering Peggy on Lakebed 2/23/2021 Demonstrating robustness to crash damage Haomiao Huang 8

Sliding Turn Post-processed analysis of a single sliding turn. Green segments denote sliding. 2/23/2021 Haomiao Huang 9

Comparison of Headings 2/23/2021 Haomiao Huang 10

Analysis & Conclusions • The vehicle is capable of making sliding turns at speed, sensors can detect and measure sideslip • The system is squirrelly but controllable; after initial learning curve simple control inputs could control vehicle • Left and right turn dynamics differ substantially due to nonlinearities in the transmission gearing, but can be adjusted for by steering input • Conclusion: Autonomous control of the vehicle in sliding turns will be nontrivial but achievable 2/23/2021 Haomiao Huang 11

Plans for Future Work: Autonomous Driving Steering sequence derived from logged human control inputs Motion primitive derived from matching trajectory to steering inputs Use primitives as feedforward term to linearize the dynamics of the system 2/23/2021 Haomiao Huang 12

Acknowledgments Many thanks to: - Prof. Richard Murray - Lyle Chamberlain - Dmitry Kogan - Mike the Shop Dude - John van Deusan - Tully Foote 2/23/2021 Haomiao Huang 13