Wayve researcher: robotics unlikely to find objectives far beyond next-token prediction
m_wulfmeier · x · 2026-09-15
Amid recent discussion on better training objectives, mwulfmeier argues from a robotics data-efficiency angle that nothing convincingly better than next-token likelihood has emerged: multi-token prediction, offline RL, and inverse RL are all weighted variants of predicting future tokens, and no alternative is both computationally cheap and robust at scale.
More from Embodied
- Physical Intelligence unveils OM-1, a robot foundation model trained purely on human data — zipengfu · 2026-09-15
- Eren Chen launches robot hardware shop: Agibot humanoids from $29,999, US shipping — chris_j_paxton · 2026-09-15
- OpenAI acquires camera startup Glass Imaging for over $300M, WSJ reports — rohanpaul_ai · 2026-09-15
- Amid new robotics launches, a pointer to best practices for policy evaluations — eigenron · 2026-09-15
- "Learning robotics today is like learning to code in 2010" — Paimaamu · 2026-09-15
- Mecha Corp unveils Spike, an RL-trained bipedal robot running purely on proprioception — Scobleizer · 2026-09-15