Robust Terrain-Adaptive Vehicle Planning and Control

Autonomous or teleoperated navigation of unmanned ground vehicles (UGVs) is difficult even in benign environments due to challenges associated with perception, decision making, and human-machine interaction, among others. In environments with rough, sloped, slippery, and/or deformable terrain, the difficulty of the navigation problem increases dramatically. In this effort, Quantum Signal, LLC, University of Michigan, and Massachusetts Institute of Technology propose to collaboratively research methods for robust terrain-adaptive planning and control to enable a future generation of UGVs with assured mobility in highly challenging terrain. The approach will exploit physics-based terrain modeling with data-driven variance estimation, stochastic vehicle motion planning through feasible corridor, and terrain-adaptive predictive vehicle control integrated into a threat-based control arbitration architecture. This architecture will enable operation at (and seamless transition between) any point on the autonomy spectrum, ranging from manual teleoperation to full autonomy. In Phase 1 the team will develop, test, and characterize algorithm performance with Quantum Signal"s high fidelity ANVEL robotic vehicle simulator and determine feasibility. Should the methods prove feasible, Phase 2 will involve the further development, integration, and testing of the methodology on experimental vehicle hardware.