Reproduce “A Human-in-the-Loop Confidence-Aware Failure Recovery Framework for Modular Robot Policies” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rohan Banerjee, Krishna Palempalli, Bohan Yang, Jiaying Fang, Alif Abdullah, Tom Silver, Sarah Dean, Tapomayukh Bhattacharjee
Cornell University researchers developed a human-in-the-loop framework for modular robot policies that optimally balances task success with human workload during failure recovery. The system systematically decides which module to query and when to solicit human input, demonstrating improved recovery efficiency and user satisfaction in robot-assisted feeding tasks.
Key insights
- Optimally balancing task success with human workload requires explicit consideration of both calibrated module confidence and the predicted cognitive/physical cost of human intervention.
- Graph-based module selectors, like GraphQuery, effectively identify the failing module by considering both uncertainty and workload, outperforming confidence-only heuristics, especially in heterogeneous systems.
- Systematic evaluation through synthetic and real-world experiments is crucial for understanding the trade-offs of different failure recovery strategies across varying system configurations and user characteristics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.