DiMuRe: Diagnosis-Guided Multilevel Recovery for Robotic Task Execution
DiMuRe: Diagnosis-Guided Multilevel Recovery for Robotic Task Execution
This research presents DiMuRe, a framework enabling robotic agents to recover from disturbances by selecting between program reuse and predefined recovery skills using phase-specific evidence. The system utilizes a multimodal large language model (MLLM) to generate corrective code after the original program exits, with local verification ensuring safety. Evaluated on 801 disturbed cases in the LIBERO-Disturb benchmark, Local DiMuRe achieves a 77.65% complete-task success rate compared to 58.68% for the original program, while Full DiMuRe reaches 81.90% with terminal replanning.
The paper introduces DiMuRe, a framework combining program reuse and corrective code generation via MLLMs to handle mid-execution disturbances, validated on the LIBERO-Disturb benchmark with reported success rates of 77.65% (Local) and 81.90% (Full).
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
program-based agents and VLA policies. On 801 scheduled disturbed Encore development cases, Local DiMuRe (program reuse and recovery skills) achieves 77.65% complete-task success, compared with 58.68% for the original program under the same physical-step allowance. Full DiMuRe achieves 81.90% with terminal replanning, an additional step allowance, and early handoff. Local DiMuRe also uses fewer physical steps per sc
Open source S4
heduled episode than the original program.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org