ROBOTICS HARNESS OPTIMIZATION ON GRAPH-AS-POLICY

Summary from the TLDR Data newsletter

Robotics Harness Optimization used coding agents to evolve a Graph-as-Policy robot controller, boosting simulated throughput 5.27x and success from 67% to 96%. The strongest version only rewired existing skills, without editing them or using human demonstrations.