Matteo Merler, Bowen Li, Josh Roy, Yichao Liang, Qianwei Wang, Yixuan Huang, Tom Silver
This paper investigates the use of coding agents for generalized task and motion planning problems, finding that they outperform hand-engineered planners and other baselines. The agents synthesize programs that generalize across instances, reducing planning effort on new instances. They achieve high success rates with less computation.
The research has practical significance as it can automate the process of task and motion planning, reducing the need for substantial engineering effort. This can lead to more efficient and effective planning in various applications, such as robotics and autonomous systems. The findings can also inform the development of more advanced coding agents for complex planning problems.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.