Robot "physics problem" stems from training data, not the model, says researcher
chris_j_paxton · x · 2026-11-03
chrisjpaxton argues the recurring "physics problem" in robotics is a training data issue, not a model one: robot training data is usually too narrowly focused on successes to build a general simulator. He notes this is conceptually easy to fix (though not practically) and that people are actively working on it — while stressing this isn't a criticism of the underlying work, which he calls super cool.
More from Embodied
- Ropedia week in review: how human first-person data becomes real robot capability — liuziwei7 · 2026-11-04
- Walden Robotics: full production shifts in factories are shaping its robot development — adnothing · 2026-11-04
- Imperial College team raises $40M for graph-generation fast imitation learning — k7agar · 2026-11-04
- Hugging Face's Reachy Mini robot just got a hoverboard — huggingface · 2026-11-03
- University of Chicago unveils RingIO: an eyes-free ring with touch as both input and output — plopesresearch · 2026-11-03
- Robotics team preps live customer demo: 1.5 days to collect data and train on-site — DominiqueCAPaul · 2026-11-03