RHO: Your Coding Agent is Secretly a Roboticist

Lifelong robot-manipulation agents must keep improving across changing objects, scenes, task phrasings, and perturbations over long deployments, while staying inspectable and reusable. Code-as-Policies (CaP) has shown that large language models (LLMs) can write code to solve robotics tasks by composing perception, planning, and control primitives. Recent CaP systems, however, rely on multi-turn code-generation loops at test time, which is often infeasible for real-time robot control. We introduce Robotics Harness Optimization (RHO), a novel paradigm in which tool-enabled coding agents, at training time, propose and search for interpretable, neurosymbolic multi-file policy repositories (Repositories-as-Policies) that compose these primitives rather than a single prompt, function, or file. RHO searches with reflective feedback from environment reward and execution rather than teleoperation demonstrations. Because a Repository-as-Policy is a persistent, version-controlled neurosymbolic artifact grown over generations of reflective search, it doubles as a substrate for training-time self-improvement that both agents and humans can inspect and extend. It generalizes to perturbed pick-and-place settings like LIBERO-PRO, where OpenVLA scores 0.0% and π0.5 averages 12.83%. Using the same low-level primitives, RHO reaches a 45.0% success rate, 2.5× higher than the strongest multi-turn agentic system, and 3.5× higher than π0.5. On Robosuite, RHO sets a new state-of-the-art of 70.0%, exceeding the prior multi-turn record of 68.29% using single-turn execution with no corrective LLM code edits at deployment. When an LLM is used in the control loop, as on RAI’s O3DE benchmark, RHO optimizes the deployed agent’s multi-file harness of prompts, tools, and control code, improving held-out success from 23.5% to 44.3% with 20% less wall-clock time and 27% fewer tool calls.