Publications

• Sorted by Date • Classified by Publication Type • Classified by Research Category •

Balancing the Mental Load: Adaptive Human-Agent Approaches for Peak Performance

Deborah van Sinttruije, Frans A. Oliehoek, and Catharine Oertel. Balancing the Mental Load: Adaptive Human-Agent Approaches for Peak Performance. In Proceedings of the AAMAS Workshop on Adaptive Learning Agents (ALA), May 2026.

Download

(unavailable)

Abstract

In high-stakes environments, where errors have severe consequences, designing adaptive systems that adjust in real-time to a user’s cog- nitive state is valuable but challenging due to the non-observable nature of these states. This study explores a partially observable Markov decision process (POMDP) framework to infer hidden cog- nitive states and dynamically manage cognitive load, minimizing the risk of cognitive overload. We tested two models: one based on established literature and another fine-tuned with user data using the Bayesian Particle Marginal Metropolis-Hastings (PMMH) method. Both models were evaluated against a performance-adaptive baseline in a user study. Our findings show that POMDP-based agents significantly reduce errors, improve task performance over time, and provide a more balanced perceived task difficulty. These results suggest that while POMDP-based adaptive systems can improve human performance, future work on cognitive adaptive systems should focus on refining model estimation techniques to better capture individual cognitive states.

BibTeX Entry

@inproceedings{Sinttruije26ALA,
    title =     {Balancing the Mental Load: Adaptive Human-Agent Approaches for Peak Performance},
    author =    {van Sinttruije, Deborah and Frans A. Oliehoek and Catharine Oertel},
    booktitle = ALA,
    year =      2026,
    month =     may,
    OPTurl =       {},
    keywords =  {refereed, workshop},
    abstract={
    In high-stakes environments, where errors have severe consequences,
    designing adaptive systems that adjust in real-time to a user’s cog-
    nitive state is valuable but challenging due to the non-observable
    nature of these states. This study explores a partially observable
    Markov decision process (POMDP) framework to infer hidden cog-
    nitive states and dynamically manage cognitive load, minimizing
    the risk of cognitive overload. We tested two models: one based
    on established literature and another fine-tuned with user data
    using the Bayesian Particle Marginal Metropolis-Hastings (PMMH)
    method. Both models were evaluated against a performance-adaptive
    baseline in a user study. Our findings show that POMDP-based
    agents significantly reduce errors, improve task performance over
    time, and provide a more balanced perceived task difficulty. These
    results suggest that while POMDP-based adaptive systems can
    improve human performance, future work on cognitive adaptive
    systems should focus on refining model estimation techniques to
    better capture individual cognitive states.
    }
}

Generated by bib2html.pl (written by Patrick Riley) on Thu May 21, 2026 16:29:56 UTC