Institution profile

Ningbo Institute of Digital Twin

Academic institutionasia · cn
Research library6linked papers
Opportunities0open roles
Selected work

Representative Papers

HygieneRoboBench: Benchmarking Hygiene-Aware Planning for Household Robots

Oct 06, 2026

This study addresses the overlooked risks of contact contamination and plan invalidation caused by novel contact events in domestic robot planning. To this end, it proposes Hygiene-NSP, a method that jointly optimizes hygiene management and task execution. Technically, the approach integrates large language model grounding, contact history reconstruction, and a CP-SAT solver to achieve hybrid planning. Additionally, a benchmark is constructed to systematically evaluate the planner’s ability to balance hygiene risk identification, cost control, and user preferences. Experimental results demonstrate that the proposed method attains a safe completion rate of 94.4% and an optimal safety rate of 90.4%, significantly outperforming existing baselines. These findings indicate that Hygiene-NSP overcomes the limitations of conventional planners that fail to simultaneously ensure safety and cost optimality.

0 citationsRead paper

Serve Now or Improve Later? Scheduling Self-Evolution in Online Agent Systems

Oct 05, 2026

This study addresses the GPU resource contention between online agent self-evolution and real-time serving, which causes delayed returns and high recovery costs. We propose LearnSched, a scheduler that incorporates capability reuse windows and computation regression times into a unified investment model to dynamically determine evolution actions. Methodologically, we construct a state-aware scheduling framework integrating candidate progress, checkpoint overhead, and recovery paths for global optimization, alongside a one-step counterfactual reasoning algorithm to select optimal execution strategies. Experimental results demonstrate that our approach significantly improves net system value under cold-recovery scenarios, validating that preserving recoverable states effectively shortens capacity regression cycles and reduces scheduling complexity.

0 citationsRead paper
Recent publications

Latest Papers

HygieneRoboBench: Benchmarking Hygiene-Aware Planning for Household Robots

Oct 06, 2026

This study addresses the overlooked risks of contact contamination and plan invalidation caused by novel contact events in domestic robot planning. To this end, it proposes Hygiene-NSP, a method that jointly optimizes hygiene management and task execution. Technically, the approach integrates large language model grounding, contact history reconstruction, and a CP-SAT solver to achieve hybrid planning. Additionally, a benchmark is constructed to systematically evaluate the planner’s ability to balance hygiene risk identification, cost control, and user preferences. Experimental results demonstrate that the proposed method attains a safe completion rate of 94.4% and an optimal safety rate of 90.4%, significantly outperforming existing baselines. These findings indicate that Hygiene-NSP overcomes the limitations of conventional planners that fail to simultaneously ensure safety and cost optimality.

0 citationsRead paper

Serve Now or Improve Later? Scheduling Self-Evolution in Online Agent Systems

Oct 05, 2026

This study addresses the GPU resource contention between online agent self-evolution and real-time serving, which causes delayed returns and high recovery costs. We propose LearnSched, a scheduler that incorporates capability reuse windows and computation regression times into a unified investment model to dynamically determine evolution actions. Methodologically, we construct a state-aware scheduling framework integrating candidate progress, checkpoint overhead, and recovery paths for global optimization, alongside a one-step counterfactual reasoning algorithm to select optimal execution strategies. Experimental results demonstrate that our approach significantly improves net system value under cold-recovery scenarios, validating that preserving recoverable states effectively shortens capacity regression cycles and reduces scheduling complexity.

0 citationsRead paper