Keep pulling the thread on Jim Fan.
On a set of 8 new Variational Automation (VA) benchmarks, the Graph-as-Policy (GaP) system achieved success rates that significantly outperformed baselines.
Graph-as-Policy (GaP) is a multi-agent coding harness that generates directed computation graphs with perception, planning, and control nodes from a Modular Open Robot Skill Library (MORSL).
The Graph-as-Policy (GaP) system uses an internal simulation environment to rehearse tasks in parallel, iteratively refining its graph structure and parameters to improve success rates and throughput.
The Graph-as-Policy (GaP) system was evaluated on 8 new open Variational Automation (VA) task benchmarks, which included 4 in-simulation and 4 real-world scenarios.