RATs: robots use Gemini to invent their own play tasks and build reusable code skill libraries
DynamicWebPaige · x · 2026-11-10
Highlighted at a conference by Google's Paige Bailey, RATs (Playful Agentic Robot Learning) lets robots autonomously propose intrinsic goals, practice them via Code-as-Policy, and distill successful behaviors into a reusable code skill library for later held-out tasks. A companion paper, GaP (Graph-as-Policy), uses Gemini 3.1 Flash-Lite as both coding agent and VLM backbone to synthesize and self-refine executable computation graphs for manipulation — showing lightweight models shining at robot control.
More from Embodied
- Newton: An open-source GPU-accelerated physics engine built on NVIDIA Warp — ChongZzZhang · 2026-11-11
- From quantum to AI: Ge Yang on CoRL's boom and where robot learning goes next — EpisodeYang · 2026-11-11
- Newton-FBA ships sim2real cloth manipulation, learned in minutes — ChongZzZhang · 2026-11-11
- Perceptron open-sources EgoHome, a sub-second egocentric household-video benchmark — lukas_m_ziegler · 2026-11-11
- Anthropic touts Fable 5.2 robot control gains on RoboDojo; real-world box folding disappoints — _Stocko_ · 2026-11-11
- AI2 open-sources ToyBlocks to benchmark robotic world models by task-relevant fidelity — allen_ai · 2026-11-11