Coding Agents Outperform Hand-Engineered Planners in Task and Motion Planning
程式碼生成 Agent 突破機器人規劃瓶頸：自動編寫高泛化性的 TAMP 解決方案
Task and Motion Planning (TAMP) is traditionally limited by hand-engineered rules. This study evaluates Claude Code (Opus 5) and Codex (GPT-5.6 Sol / GPT-6 Astra) on synthesizing generalized TAMP programs across 28 simulator environments. By interacting with simulations to calibrate strategies, the agents generated code evaluated across 98,000 episodes. The resulting programs achieved 56% to 95% success rates—outperforming classical planners (47%)—and maintained high efficiency as object scales increased, reducing computational cost by an order of magnitude.
Key points
Interactive Program Synthesis
Agents leverage interactive simulator access to calibrate physical models, test edge cases, and refine strategies within a fixed synthesis budget.
Outperforming Classic Planners
Across 16 environments, all agent configurations beat traditional hand-engineered planners, scoring 56%-95% mean success versus the planners' 47%.
Scalability to Higher Complexity
As object counts scale up, the synthesized programs maintain robust success while requiring an order of magnitude less computation per instance.
How it works
Why it matters
This research proves that coding agents can automate complex robotic planning without expert hand-engineering. The 'interactive synthesis and freeze' pipeline yields highly generalizable control programs, reducing real-time computational overhead and defining a strong baseline for generalized TAMP.
Who it affects
- AI Developer
- AI Researcher
- Enterprise Leader
How to use it
- 1Automating task and motion planning code generation for robots operating in highly complex, multi-object environments.
- 2Autonomous physical model calibration and edge-case testing within dynamic simulated environments.
Limitations & caveats
- Highly reliant on active simulator access for interactive code calibration and feedback.
- Program synthesis is constrained within a fixed interactive token and computation budget.
Related
AD-WM: Action-Discriminative World Models for Counterfactual MPC
AD-WM：專為反事實預測控制設計的動作辨識世界模型
AD-WM enhances latent world models by learning to distinguish alternative actions from the same state, significantly improving counterfactual control and zero-shot transfer in robotics.
RAPID: Robot Agentic Programming from Demonstrations
RAPID：只需單次視覺示範，AI 代理即可自動生成與優化機器人操控程式
RAPID is a framework that automatically infers task specifications and environments from a single visual demonstration, using an agentic loop to program and refine generalized robot skills.
Rolling-WAM: Accelerating Robotic World Action Models via Rolling Imagination
Rolling-WAM：利用滾動想像實現高效控制的機器人世界動作模型
Rolling-WAM accelerates world action models by distributing the joint video-action denoising process across sliding windows over successive replanning cycles, achieving a 4.5x speedup.