Yacine trains robot stop-and-go policy in his own sim, sim2real in under 2 minutes
yacineMTB · x · 2026-09-01
Yacine (ex-OpenAI) demos his robot RL pipeline: training a stop-and-go policy in a simulator he wrote himself, then transferring it to MuJoCo — the whole training finishes in under 2 minutes. Everything runs on his own infrastructure, monitored from a data-checking app on his phone.
He also takes a firm methodological stance: unlike others' trainers, his is symmetric — he argues asymmetric actor-critic is "bloat", prevents gradients from flowing, and is simply bad practice ("I will die on this hill").
Related event: Asymmetric Actor-Critic Architecture Draws Criticism in Robot RL(2 posts)→
More from Embodied
- Peking University's SimForcing distills simulation motion priors into robot world models — PekingUniversity · 2026-10-16
- NVIDIA and Mercedes-Benz Expand Partnership on Next-Gen Automated Driving — nvidia · 2026-10-16
- ~25,000 Humanoid Robots Shipped in H1 2026, Over 95% from Chinese Vendors — ingliguori · 2026-10-16
- openpilot brings stoplight-stopping autonomy to a 2019 Corolla — k7agar · 2026-10-16
- Hironobu Fujiyoshi publishes deck: from AI agents to Agentic Robotics — HirokatuKataoka · 2026-10-16
- CoDA co-evolves simulation and robot policies via shared failure memory, lifting OOD success from 5.3% to 27.6% — PLAN-Lab · 2026-10-16