Efficient Trajectory Optimization for Robot Motion Planning

Planning time optimal motion with limited joint speed, acceleration, torque, or even torque rate for robots with high degrees of freedom and complicated dynamics is challenging. In this work, an efficient trajectory optimization framework that utilizing pseudospectral theory and automatic differentiation is presented to plan a time optimal, collision-free, dynamical feasible trajectory within several seconds (in the video, 3s). The resulting result has avoided non-smoothness in lots of sampling based path planning algorithms, as well as the sub-optimality caused by path-velocity decomposition approaches.

Details see publication: 'Efficient Trajectory Optimization for Robot Motion Planning', Yu Zhao, Hsien-Chung Lin, Masayoshi Tomizuka, ICARCV 2018.