Generalist model, not a robot policy, teaches itself darts and lands 7 straight shots
ZoubinGhahrama1 · x · 2026-10-14
A demo shared by TheHumanoid Hub shows a generalist model (referred to as Gemini, version unconfirmed) — not a dedicated robot policy — completing a physical control task from high-level instructions via a four-step process, landing 7 darts in a row:
- Watched a human throw and mapped the motion to its own joints;
- Used cameras to localize the board in 3D and built a physics simulation;
- Practiced in simulation, tuning stance and swing for a smooth arc;
- On the real robot, adjusted by centimeters based on each landing until 7 shots went in.
It's a notable demonstration of the "generalist model + reasoning + sim tools" approach to embodied control, rather than end-to-end robot policies.
More from Embodied
- NAVER's WATCH: Cross-Video Human-Video Pretraining Lifts Robot Success Rate to 77.7% — naver-ai · 2026-10-14
- Robot Dog Tries to Climb Its Own Back: Conitzer's Latest Funny AI Fail — conitzer · 2026-10-14
- This Week in New Models: RoboJEPA, Long-WAM, VioLA and 4 More to Know — TheTuringPost · 2026-10-14
- XPENG's AUGUstus VLA model hits SOTA on NAVSIM and deploys on real vehicles at 12Hz — XPENG-AI · 2026-10-14
- EMHO: Self-Evolving Agent Harness Boosts Frozen Embodied Models via Experience Traces — Hyun Jung Lee · 2026-10-14
- Robotics startup Origin teases 'major announcement' with robot arm image — vaibhavbetter · 2026-10-14