GPT-6 Astra Trains a Robot Dog to 9 Motions in 25 RL Rounds Over 5 Days
TheMoonMidas · x · 2026-09-12
Japanese developer @gclueakira used GPT-6 Astra over 5 days and 25 iteration rounds to teach a robot dog 9 motions via reinforcement learning in Isaac Lab simulation. Astra also designed the body in Fusion360 — QDD motors are RS05, battery is Makita 40V, board planned as Jetson — fixing issues raised each round.
Astra generated the STL files; after 5 more rounds of dimension fitting, the plan is to deploy to the physical robot, with humans executing deployment under Astra's instructions.
More from coding & agent
- 100-Question Forecasting Benchmark: Arise PiT Search Pits Deep Research Agents vs Prediction Markets — CShorten30 · 2026-09-12
- Microsoft's APM: an open-source package manager for AI agent skills, 3.8k stars on GitHub — aronchick · 2026-09-12
- DSPy 3.4.0b1 beta ships shared LM engine interface, faster execution and async ReActV2 — dbreunig · 2026-09-12
- Ex-Cursor engineer runs 10-20 Grok cloud agents in a loop: Chief of Staff, PM and 15+ workers — NicoVerderosa · 2026-09-12
- Open-source Plano routes LLM calls by prompt intent, no agent code changes, cutting bills 2x — Roger_M_Taylor · 2026-09-12
- Netflix Engineer Open-Sources Headroom, Cuts Agent Token Use by Up to 95% — Roger_M_Taylor · 2026-09-12