Researchers introduce NIABench, a simulation benchmark and LLM-based framework for non-intrusive robot assistance that proactively supports humans without interrupting their ongoing activities.

A research team has established NIABench, a simulation benchmark with new metrics designed to evaluate non-intrusive assistance—where a robot proactively supports a human's multi-step activities while strictly avoiding interruptions.

The proposed method combines an LLM (an AI that understands and generates text) with a scoring model; the scoring model uses semantic retrieval to reduce candidate action sets, then ranks human-step and robot-action pairs to reason about timing and dependencies.

Experiments on both the simulation benchmark and real-world scenarios show the method achieves proactive, non-intrusive assistance that reduces human effort while preserving task effectiveness.