Control and planning for vehicles with uncertainty in dynamics

Abstract

This paper describes a motion planning algorithm that accounts for uncertainty in the dynamics of vehicles. This noise is a function of the type of controller employed on the vehicle and the characteristics of the terrain and can cause the robot to deviate from a planned trajectory and collide with obstacles. Our motion planning algorithm finds trajectories that balance the trade-off between conventional performance measures such as time and energy versus safety. The key is a characterization of the vehicle's ability to follow planned paths, which allows the algorithm to explicitly calculate probabilities of successful traversal for different trajectory segments. We illustrate the method with a six-legged Rhex-like robot by experimentally characterizing different gaits (controllers) on different terrains and demonstrating the hexapod navigating a multi-terrain environment.