RHO: Your Coding Agent is Secretly a Roboticist

2026-06-15

AI Summary

Existing code-generation-based robotic control methods rely on multi-round generation at test time, which struggles to meet real-time requirements. This work proposes the Robotics Harness Optimization (RHO) paradigm, which leverages a tool-augmented coding agent during training to search an interpretable, neuro-symbolic, multi-file policy repository. RHO composes perception, planning, and control primitives into deployable strategies that execute efficiently in a single pass. It introduces, for the first time, a multi-file policy repository as the policy representation, enabling self-reflective optimization through environmental rewards and execution feedback—without requiring human demonstrations or multi-round interaction. Experiments show that RHO achieves a 45.0% success rate on LIBERO-PRO (a 2.5× improvement over the strongest multi-round system), sets a new state of the art at 70.0% on Robosuite, and boosts performance on the RAI O3DE benchmark from 23.5% to 44.3%, while reducing runtime by 20% and tool calls by 27%.

Abstract

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. It generalizes to perturbed pick-and-place settings like LIBERO-PRO, where OpenVLA scores 0.0% and $π_{0.5}$ averages 12.83%. Using the same low-level primitives, RHO reaches a 45.0% success rate, 2.5x higher than the strongest multi-turn agentic system, and 3.5x higher than $π_{0.5}$. 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.

Problem

• real-time robot control
• code generation
• multi-turn loops
• robotics tasks
• LLM-based control

Innovation

• Robotics Harness Optimization
• Repositories-as-Policies
• neurosymbolic policies
• single-turn execution
• tool-enabled coding agents