Efficient dynamic programming for high-dimensional, optimal motion planning by spectral learning of approximate value function symmetries
We demonstrate how to find high-quality motion plans for high-dimensional holonomic systems efficiently using dynamic programming in a learned subspace of vastly reduced dimension. Our approach (SLASHDP) learns the lowdimensional cost structure of an optimal control problem via an efficient spectral method. This structure results in a symmetric value function that serves as a an efficientlycomputable surrogate for the true value function. High-quality feedback motion plans can then be obtained from the symmetric value function. Experimental results show that SLASHDP yields higher-quality plans than can be obtained by post-processing plans generated by a sampling-based motion planner, and with less computational effort for very high-dimensional problems. We demonstrate high-quality dynamic programming plans for an arm planning problem of up to 144 dimensions without using any domain-specific knowledge aside from that learned automatically by SLASHDP. Positive results are also shown for a high-dimensional deformable robot planning problem.
Paul Vernaza Daniel D. Lee
GRASP Lab,University of Pennsylvania,3330 Walnut St.,Philadelphia,PA 19104
国际会议
2011 IEEE International Conference on Robotics and Automation(2011年IEEE世界机器人与自动化大会 ICRA 2011)
上海
英文
6121-6127
2011-05-09（万方平台首次上网日期，不代表论文的发表时间）