When to Act: Calibrated Confidence for Reliable Human Intention Prediction in Assistive Robotics

TL;DR

This paper presents a calibrated confidence framework for assistive robotics that improves safety by ensuring predictions are reliable before acting, using post-hoc calibration and a simple ACT/HOLD decision rule.

Contribution

It introduces a post-hoc calibration method for model confidence and a safety-critical triggering framework for reliable human intention prediction in assistive devices.

Findings

Calibration reduces miscalibration by about an order of magnitude.

Calibrated confidence enables a safety threshold for actions.

Framework improves reliability without sacrificing accuracy.

Abstract

Assistive devices must determine both what a user intends to do and how reliable that prediction is before providing support. We introduce a safety-critical triggering framework based on calibrated probabilities for multimodal next-action prediction in Activities of Daily Living. Raw model confidence often fails to reflect true correctness, posing a safety risk. Post-hoc calibration aligns predicted confidence with empirical reliability and reduces miscalibration by about an order of magnitude without affecting accuracy. The calibrated confidence drives a simple ACT/HOLD rule that acts only when reliability is high and withholds assistance otherwise. This turns the confidence threshold into a quantitative safety parameter for assisted actions and enables verifiable behavior in an assistive control loop.