RoboArchitect automates robotic control harness design boosting task success
RoboArchitect: Automating the Design of Robotic Agents with Persistent Memory
RoboticsArtificial IntelligenceComputer Vision and Pattern Recognition
Summary
Robots often use fixed control programs, or policies, which don’t adapt well to new jobs or environments. The paper shows how RoboArchitect uses a coding model to automatically build and improve a surrounding control framework, called a harness, that guides the robot’s actions and decisions. This means the robot can succeed more often on tasks it hasn’t seen before without changing the core policy. RoboArchitect also creates simple memory systems within the harness to remember useful information across steps. This approach improves robot manipulation by optimizing control logic rather than retraining the policy.
What this means in practice
- •For robotics engineers: Automatically generate and refine control code around fixed robot policies to improve performance on new tasks and environments.
- •For industrial automation teams: Deploy robots that adapt better in manufacturing or logistics by evolving control harnesses without retraining core policies.
Authors
Shengguang Wu, Suning Huang, Cherie Ho, Ria Doshi, Hao Zhu, Serena Yeung-Levy
Abstract
Recent work shows that a frozen robot policy performs far better inside an agent harness, where control code decides when to query the policy, consult vision tools, or command the arm directly. Yet today these harnesses are built by hand and redesigned for each new robot or benchmark. We show that harness design can be automated, offering a strong alternative to finetuning on a novel setup. In RoboArchitect, a coding model observes a robot's rollouts, diagnoses failures, and iteratively revises the harness around the frozen policy. The harness evolves on a set of episodes for its optimization and is evaluated on held-out scenes and tasks. Without updating any model weights, RoboArchitect raises average success from 9.7% to 43.3% on unseen scenes (4.5x) and from 12.7% to 40.2% on unseen tasks (3.2x), across multiple benchmarks and policies. The gains are largest where the benchmark is far from the policy's training data: wrapping a general-purpose policy, the evolved harness yields a specialist that often outperforms even the finetuned version. Intuitive control strategies are automatically designed, such as lifting a jammed arm clear or reaching in horizontally when an obstacle blocks the way from above. The harness also evolves its own persistent memory, carrying learned experience, such as effective grasp depths, into later decisions. These results suggest that substantial headroom in robotic manipulation lies in the agent harness, and its design can be effectively automated and optimized without weight updates.