Coding agents improve complex task and motion planning in robotics
Coding Agents for Generalized Task and Motion Planning Problems
RoboticsArtificial Intelligence
Summary
Planning tasks that involve both choosing actions and dealing with physical constraints is very hard for robots. The authors studied if computer programs, called coding agents, can write their own plans that work well across many different situations. They tested these agents on lots of simulated tasks and found their generated plans often did better than hand-made planners, especially for bigger and more complex problems. These agents even learn by interacting with the environment to improve their strategies. This shows coding agents can help automate and scale planning for robots.
What this means in practice
- •For robotics software developers: Automatically generate efficient task and motion plans that adapt well when objects or environments change.
- •For industrial automation engineers: Reduce computation time needed for generating robot task plans in complex, changing assembly tasks.
Authors
Matteo Merler, Bowen Li, Josh Roy, Yichao Liang, Qianwei Wang, Yixuan Huang, Tom Silver
Abstract
Task and motion planning (TAMP) problems remain difficult even with full observability and object-centric states because discrete decisions are tightly coupled to geometric, kinematic, and dynamic constraints. Generalized TAMP addresses this difficulty by exploiting regularities across problem instances to reduce planning effort on new instances. However, existing methods require substantial TAMP-specific engineering. We investigate whether coding agents can automate this process by synthesizing programs that generalize across instances. Given a task description and simulator access, each agent chooses how to interact with the environment while developing a program within a fixed synthesis budget. The program is then frozen and evaluated on unseen instances. We evaluate Claude Code (Opus 5) and Codex (GPT-5.6 Sol and GPT-6 Astra) on 28 simulated environments from KinDER and PDDLStream, with object counts beyond those evaluated in the original benchmark. Across all program synthesis methods, we evaluate 980 generated programs on 100 held-out instances each, 98,000 evaluation episodes in total. Overall, we find that coding agents are surprisingly effective at generalized TAMP: all three agent configurations outperform hand-engineered planners, one-shot generation, and an LLM-based generalized planning baseline in mean success (56% to 95% versus 47% for the planners, on the 16 environments where a planner is available). As object counts grow, the agents' programs maintain higher success than the planner, using an order of magnitude less computation per instance on average. Logs show agents using interaction to calibrate physical models, test edge cases, and refine strategies. We release all code, including the full prompts given to the agents. These findings suggest that coding agents are a strong baseline for generalized TAMP.