RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes
RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes
RoboAware is a new method for coordinating embodied skills in robots by learning from counterfactual outcomes. It uses a hierarchical MDP framework and introduces State-Locked Counterfactual Branching (SCB) and Execution-Aware Learning (EAL) to improve policy selection. The method achieves a 77.0% overall success rate and outperforms existing baselines on several robotic benchmarks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
tasks
100
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
k. Comprehensive single-episode evaluations on 100 tasks show that RoboAware reaches a 77.0% overall success rate, with SOTA averages of 90.0% on RoboSuite, 73.8% on diverse LIBERO-Pro task clusters, and 90.0% on challenging RoboTwin bimanual tasks, outperforming existing code-as-policy and VLA-harness baselines.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org