Embodied AI Study: Robot Foundation Models Consistently Fail at Chained Tasks
rohanpaul_ai · x · 2026-08-14
A new study highlights a "Knowing, but not Noticing" flaw in current Embodied Robotics (ER) models. The research shows that frontier ER models—which robot foundation models predominantly initialize from—consistently fail to know 'what to do next' when executing chained tasks.
The team proposes a method to quantitatively measure this failure using data sourced from a novel branching, 'counterfactual'-style ego data collection regime. They also introduce a means to produce post-training data to fill these gaps, extracting context the model already knows but fails to conjure reflexively on its own via context ablations through harnessing.
More from Embodied
- FoveaVLA replaces wrist cameras with learned foveation, hitting 70.8% success — Haoyu_Xiong_ · 2026-10-16
- Musk teases two-way Grok x Optimus integration: Grok in the bot, bot in Grok — beffjezos · 2026-10-16
- SimVTA uses simulator-rendered video to post-train driving world action models with RL — OpenDriveLab · 2026-10-16
- Robotics data startups compared: Scale AI tops $1B revenue as 9 LLM vs robotics data suppliers benchmarked — FinanceYF5 · 2026-10-16
- Kroger cuts 3,000 corporate jobs (37%) for AI; humanoid robots next in the 2030s — csuwildcat · 2026-10-16
- Tesla dares its humanoid robot to do a kickflip in playful demo — yunta_tsai · 2026-10-16