Embodied Skill Bench: Benchmarking How Well Agents with Embodied Skills Perform on Physical Tasks
Embodied Skill Bench: Benchmarking How Well Agents with Embodied Skills Perform on Physical Tasks
This research introduces EmbodiedSkillBench, a framework that evaluates the effectiveness of embodied skills in enhancing robotic agents' performance on physical tasks. The study demonstrates that skills improve performance for weaker agents but offer diminishing returns as agents become more capable. The ADR mode allows agents to selectively use skills while retaining native control, balancing performance and efficiency.
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.
- code: Available Artifact link Open source S6
- code: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
nd decides for each skill whether to invoke, modify, or ignore it. Across four agents on 100 perturbed LIBERO-Pro tasks, we find that skills help an agent only as far as they exceed what it can already do. Skill-only execution lifts GPT-5.6-Sol from 11% to 63% and Kimi-K3 and Claude-Opus-5 by about 20 points, while cutting their token usage by 58% to 85%. Yet all four agents then end up between 61% and 64%, and GPT-6
Open source S4
-Astra falls from 92% under native control to 61%. This suggests that as agents grow stronger, the role of embodied skills shifts from supplying missing capability to reducing cost: GPT-6-Astra already solves most tasks through native control, so restricting it to skills removes more than it adds. Under ADR, which lets it keep native control and call skills selectively, GPT-6-Astra recovers to 85%, within 7 points of
Open source S5
native control, while using 14.1% fewer tokens and 11.9% less time. With Jev as a skill choosing motion directions and gripper actions, it reaches 81% with 50% fewer tokens and 57% less time. The code is available at https://github.com/HanZhi306/EmbodiedSkill-Bench.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org