RL post-training can improve generalist robot policies on novel unseen tasks
arkrause · x · 2026-11-06
Michal Korniak shares a thread showing that RL post-training of generalist robot policies can deliver "generalist improvement": with the right RL recipe, policies improve not just on trained tasks but also on novel tasks never seen during training.
More from Embodied
- China's robotics flywheel: scale, cost, standards and field data — ingliguori · 2026-11-06
- Liquid AI's LFMs roll out in Mercedes-Benz vehicles, running fully on-device — JosephJacks_ · 2026-11-06
- GPT-6.1 Sol halves surgical positioning error after 3 mouse brain surgeries — 141_1337 · 2026-11-06
- Walmart to add 300 drone delivery sites via Zipline expansion, deliveries in 30 minutes — Polymarket · 2026-11-06
- Nvidia-backed Agility Robotics debuts on Nasdaq as first dedicated US humanoid robot IPO — Polymarket · 2026-11-06
- AI hardware map for 2026: 139 companies across 14 layers, biggest valuations on your wrist — FinanceYF5 · 2026-11-06