Event
RoboAuto: Multi-level Physical Understanding Drives Robot Program Search
1 reports · 1 sources
Buzhao Liu and other authors propose RoboAuto, a robot program search framework that organizes physical understanding at three levels—object, scene, and interaction—to guide LLM agents in generating and revising manipulation programs. The authors claim that the method performs better on five benchmark series: achieving an 89.1% success rate on short-horizon LIBERO-PRO tasks, and an average success rate more than twice that of ASPIRE on LIBERO-PRO Long and LIBERO-Plus. The authors also claim that RoboAuto outperforms ASPIRE and π0.5 on three long-horizon real-robot tasks requiring fine manipulation; its knowledge base can transfer across simulators and embodiments, and program search performance improves with knowledge base size. These results are all author-reported and await independent verification.
Report timeline
- RoboAuto: Autonomous Robot Program Search through Multi-Level Physical Understanding
RoboAuto proposes a robot program search framework that organizes physical understanding at three levels—object, scene, and interaction—to guide LLM agents in generating and revising manipulation programs. The method achieves better performance across five benchmark suites, reaching an 89.1% success rate on short-horizon LIBERO-PRO tasks and averaging more than twice the success rate of ASPIRE on LIBERO-PRO Long and LIBERO-Plus.
arXiv cs.RO ·