Dynamic harness improves robot coordination and self-correction during tasks
DynaHarness: A Dynamic Physical Harness for Self-Evolving Robot Agents
RoboticsArtificial IntelligenceMachine Learning
Summary
Robots often struggle with long tasks because their thinking and doing happen at different speeds, and when they fail, it’s hard to know what went wrong. The authors present DynaHarness, a system that lets a robot’s slow 'brain' plan and a fast 'brain' monitor actions closely, catching problems early. It uses a contract to track what commands are allowed and learns from failures to fix its skills. This approach helped their robot reach much higher success in tests than a fixed policy.
What this means in practice
- •For industrial robot operators: Enable robots to autonomously detect and fix execution errors during complex tasks, reducing downtime and manual intervention.
- •For robotics software engineers: Implement a dynamic control architecture that tightly couples planning and execution with failure-informed self-improvement in robotic agents.
Authors
Haoyuan Deng, Jiebin Liu, Tengxiao Zhang, Langning Yan, Hongye Cao, Ziwei Wang
Abstract
Pretrained robot policies provide useful action priors, but long-horizon manipulation still requires coordination between semantic reasoning and physical execution. Semantic reasoning operates at a coarser timescale than physical interaction, while episode-level failures provide limited guidance on which system component should be revised. We propose DynaHarness, a dynamic physical harness that couples semantic reasoning with physical governance through a shared execution contract and turns failure evidence into validated capability revisions. To be more specific, the slow brain proposes capabilities and symbolic arguments, while the fast brain grounds and monitors commands, refuses unresolved actions, substitutes capabilities, and requests replans when needed. The physical execution contract bounds each accepted command and records execution evidence across analytic skills, recovery skills, and the frozen VLA. Failure attribution localizes faults in these records and directs targeted revisions of reusable capabilities or execution mechanisms. Paired regression checks govern admission or rejection, closing the self-evolution loop. On LIBERO-Pro, DynaHarness achieves 75.2% on 800 newly sampled initial states, compared with 17.5% for the frozen policy. With the same capability library, full dynamic execution reaches 74.0% versus 63.9% under nominal one-step replanning. This demonstrates the value of DynaHarness as a dynamic physical harness that governs how existing capabilities are grounded, monitored, and coordinated during execution. Our project page is at https://denghaoyuan123.github.io/Dynaharness_page/.