High-DOF Motion Planning in Dynamic Environments using Trajectory Optimization
Chonhyon Park · Carolina Digital Repository (University of North Carolina at Chapel Hill) · 2019
Motion planning is an important problem in robotics, computer-aided design, and simulated environments. Recently, robots with a high number of controllable joints are increasingly used for different applications, including in dynamic environments with humans and other moving objects. In this thesis, we address three main challenges related to motion planning algorithms for high-DOF robots in dynamic environments: 1) how to compute a feasible and constrained motion trajectory in dynamic environments; 2) how to improve the performance of realtime computations for high-DOF robots; 3) how to model the uncertainty in the environment representation and the motion of the obstacles. We present a novel optimization-based algorithm for motion planning in dynamic environments. We model various constraints corresponding to smoothness, as well as kinematics and dynamics bounds, as a cost function, and perform stochastic trajectory optimization to compute feasible high-dimensional trajectories. In order to handle arbitrary dynamic obstacles, we use a replanning framework that interleaves planning with execution. We also parallelize our approach on multiple CPU or GPU cores to improve the performance and perform realtime computations. In order to deal with the uncertainty of dynamic environments, we present an efficient probabilistic collision detection algorithm that takes into account noisy sensor data. We predict the future obstacle motion as Gaussian distributions, and compute the bounded collision probability between a high-DOF robot and obstacles. We highlight the performance of our algorithms in simulated environments as well as with a 7-DOF Fetch arm.