Robot world model demo shows a grasp that never happened in reality
Negative-Whereas3307 · reddit · 2026-11-05
A Reddit post highlights a WorldOracle side-by-side case: the real robot missed a piece of fruit, but the model's predicted video — generated from the exact same starting observation and recorded actions — showed a plausible, successful pickup.
Key points:
- With identical inputs, the model produced a convincing video of something that didn't happen
- Without the real execution shown alongside, such predictions could pass as a perfectly reasonable account
- The author argues "good prediction" should mean faithfully predicting failure, not showing success while getting the outcome wrong
The post raises the question: how much can we judge from robot world-model demos when the real execution isn't shown for comparison?
More from Embodied
- Berkeley project uses pretrained base policy's support to guide exploration for fast online adaptation — abhishekunique7 · 2026-11-05
- Ex-robotics founder predicts startup wave: data collection and sim-as-a-service are entry points — carlosdponx · 2026-11-05
- Robotics sim platform startups are coming: data collection and sim-as-a-service — carlosdponx · 2026-11-05
- "Robotics simulation platform" startups are a dead end — agents already removed the need — carlosdponx · 2026-11-05
- Runtime (YC F26) debuts simulation-to-real robot policies, threads a rope with no fine-tuning — ycombinator · 2026-11-05
- Rooms3d turns Vision Pro room scans into permanent 3D dioramas for $9.99, one-time purchase — Scobleizer · 2026-11-05