Evolutionary search on Graph-as-Policy robot controllers yields 5.27x throughput for $37
matei_zaharia · x · 2026-09-28
Following his Robotics Harness Optimization (RHO) paper with AMD—using evolutionary search to improve robot policies that are whole code repositories—Karim applied RHO to NVIDIA's Graph-as-Policy (GaP, arXiv:2607.05369), where humans write skill nodes (observe, perceive, grasp, move, release) and an LLM wires them into a directed computation graph. After 50 generations of evolutionary search on an example graph: 5.27x throughput gain over baseline, grasps per trial down from 4.17 to 1, and total search cost of just $37 across both experiments. GaP's graph policies are readable, diffable source code with locally editable, searchable stages, letting search propose "add a node" moves without demonstrations. Matei Zaharia amplified the work.
More from Embodied
- RollerCoaster Tycoon goes spatial: 1999 pixel game rebuilt with voxels on Specs — stspanho · 2026-09-28
- RoboSSM: State-Space Models Enable Robots to Keep Improving at Test Time Without Fine-Tuning — yifengzhu_ut · 2026-09-28
- Robotics pioneers split on humanoid timelines: action data, not LLMs, is the bottleneck — Scobleizer · 2026-09-28
- Berkeley's TRACE robots trace monochrome data center cables via bidirectional tracing — berkeley_ai · 2026-09-28
- macOS 27 beta's new hit-test map debug tool fuels touchscreen MacBook rumors — stephancasas · 2026-09-28
- IFR World Robotics 2026: Five million robots now operate in factories globally — Recoil42 · 2026-09-28