RobotAgentBench debuts to test open-world robot autonomy: manipulation ≠ autonomy
HennyJieCC · x · 2026-10-28
Lingjun Mao's team introduced RobotAgentBench, a benchmark for evaluating open-world robot autonomy in realistic household environments, testing VLA/WAM policies alongside diverse robot agent harnesses.
It spans five dimensions: task discovery, context reasoning, manipulation, failure recovery, and robustness, across four open-ended household scenarios that test how these capabilities combine to fulfill high-level user needs.
Key takeaway: manipulation ≠ autonomy — standalone VLA/WAM policies struggle with task discovery and context reasoning, so truly autonomous robot assistants remain a distant goal.
More from Embodied
- New benchmark tests whether home robots navigate more cautiously around babies than trash bins — SeoulNatlUniv · 2026-10-28
- Early adopter revisits AI hardware: still meh, runs hot without opening apps — alexmacgregor__ · 2026-10-28
- Would you ship a policy trained mostly in generated worlds? AIFlow polls robot people — AIFlow_ML · 2026-10-28
- Re:Build CEO: robots attack a small slice of production cost, lean beats them — jaimalik · 2026-10-28
- Baidu's Apollo Go holds first large-scale public robotaxi test ride in Hong Kong — Baidu_Inc · 2026-10-28
- DeepMind caveat: robot grasp policies drift on object permanence, sims still win — AIFlow_ML · 2026-10-28