AgenticGenTAMP: coding agents synthesize reusable TAMP policies across 28 simulated environments
tomssilver · x · 2026-11-10
Princeton-led AgenticGenTAMP (CoRL) uses coding agents to synthesize a single programmatic policy per environment for generalized task and motion planning, tested across 28 simulated environments with varying object counts, configurations and geometry. Agents get a task description and a $20 model-usage budget per synthesis run, design their own simulator tests, and run in isolated Docker containers with no network access. Paper, code and dataset are open, with new Opus 5.5 results shown.
More from Embodied
- Pincer1, a six-legged robot hardware wallet, opens preorders with lidar room verification — R4nsack · 2026-11-10
- The Second Station: why 5x-ing one factory station doesn't save the line — saranormous · 2026-11-10
- Tesla VP Explains the In-Cabin Radar That Detects Kids and Pets Left in Cars — XFreeze · 2026-11-10
- Video tampering tricks every frontier embodied AI model into harmful acts, 100% baby-stab rate — DJiafei · 2026-11-10
- YAM robot arm picks and boxes a screwdriver in under 15s at CoRL blind-eval exhibit — DJiafei · 2026-11-10
- PsiBorg demo: robot works toward a goal image to rearrange objects autonomously — dyamins · 2026-11-10