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

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Gaus, Johannes A., Ilg, Winfried, Haeufle, Daniel
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914241186889728
author Gaus, Johannes A.
Ilg, Winfried
Haeufle, Daniel
author_facet Gaus, Johannes A.
Ilg, Winfried
Haeufle, Daniel
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2601_04982
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle When to Act: Calibrated Confidence for Reliable Human Intention Prediction in Assistive Robotics
Gaus, Johannes A.
Ilg, Winfried
Haeufle, Daniel
Robotics
Artificial Intelligence
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.
title When to Act: Calibrated Confidence for Reliable Human Intention Prediction in Assistive Robotics
topic Robotics
Artificial Intelligence
url https://arxiv.org/abs/2601.04982