Coding Agents for Generalized Task and Motion Planning Problems
Coding agents write programs that beat hand-engineered planners on generalized task and motion planning.
Researchers test whether coding agents can synthesize programs for generalized task and motion planning. Claude Code using Opus 5 and Codex using GPT-5.6 Sol and GPT-6 Astra were run on 28 KinDER and PDDLStream environments, yielding 980 programs and 98,000 evaluation episodes. All three agent configurations beat hand-engineered planners, one-shot generation, and an LLM planning baseline, with mean success of 56% to 95% versus 47% for planners on the 16 environments where a planner exists. As object counts increase, the generated programs remain more successful while using about an order of magnitude less computation per instance.
- Agents write a program from a task description and simulator access, then freeze it.
- 980 programs were evaluated on 100 held-out instances each, 98,000 episodes total.
- Mean success was 56% to 95% versus 47% for available hand-engineered planners.
- Agent programs stay more successful as object counts grow and use far less compute.
- The authors release all code and the full prompts given to the agents.
Full article265 words · extracted from huggingface.co · click to collapse
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.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.30233