|
abstract:
Review paper
Toward autonomous artificial intelligence agents in sports science:
a modular framework for development, validation, and
implementation
Ismail Dergaa
1, 2, 3
 ,
Sabri Barbaria
4
,
Wissem Dhahbi
3, 5
,
Piotr Zmijewski
6
,
Karim Chamari
7
,
Helmi Ben Saad
8, 9, 10
- High Institute of Sport and Physical Education of Ksar Said, University of Manouba, Manouba, Tunisia
- Research Unit Physical Activity, Sport, and Health, UR18JS01, National Observatory of Sport, Tunis 1003,
Tunisia
- High Institute of Sport and Physical Education of Kef, University of Jendouba, El Kef, Tunisia
- Laboratory of Biophysics and Medical Technologies, LR13ES07 (BTM), Higher Institute of Medical Technologies
of Tunis (ISTMT), University of Tunis El Manar, Tunis 1080, Tunisia
- Training Department, Police College, Qatar Police Academy, Doha, Qatar
- Jozef Pilsudski University of Physical Education in Warsaw, Warsaw, Poland
- Research Office, Naufar Center, Doha, Qatar
- Faculty of Medicine of Sousse, University of Sousse, Farhat Hached University Hospital, Laboratory of Physiology,
Sousse, Tunisia
- University of Sousse, Faculty of Medicine of Sousse, Farhat Hached University Hospital, Laboratory research
LR12SP09 “Heart failure” Sousse, Tunisia
- University Central Group, UPSAT-Sousse, CRCI, Tunis, Tunisia
Biol Sport.2026;43:1353–1426
Online publish date: 2026/05/12
PlumX metrics:
Despite widespread artificial intelligence (AI) adoption in sports science for predictive analytics, current systems operate as passive tools requiring continuous human monitoring and intervention at every decision point. Autonomous AI agent systems capable of 24/7 monitoring, independent reasoning, and proactive action execution remain academically unexplored in sports science contexts. Unlike passive analytics that await human analysis or conversational interfaces requiring explicit prompting, autonomous agents operate continuously, detecting patterns and implementing interventions without human initiation. Our review distinguished autonomous agents from existing AI applications, proposes modular implementation frameworks, develops theoretical application workflows across eight priority domains, and establishes empirical validation pathways. Current (i.e., April 23, 2026) literature lacks peer-reviewed research on autonomous agent systems in sports science. This review connects computer science with exercise physiology. We integrate modern agent architectures with established sports science concepts. The outcome is a practical, multi-domain implementation roadmap. Our three-phase framework progresses from specialized single-domain agents through coordinated multi-agent systems to fully integrated platforms. Phase 1 develops autonomous agents for training load management, exercise prescription, biomechanical analysis, nutrition optimization, sleep monitoring, injury prevention, mental skills training, and rehabilitation, each operating independently within defined safety boundaries. Phase 2 stablishes coordination protocols enabling information exchange across domains while maintaining modular independence. Phase 3 integrates fully autonomous agent systems across all domains into a unified platform with comprehensive cross-domain reasoning. This framework aimed to advance autonomous agent research from conceptual proposal to a structured implementation and validation pathway in evidence-based athlete management.
Video abstract: https://www.youtube.com/watch?v=MPL-NP8kzpo&feature=youtu.be
keywords:
AI agents, Algorithmic bias, Athletic performance, Agentic frameworks, ChatGPT, Computer vision, Exercise prescription, Human oversight, Implementation science, Injury prevention, Large Language Models, Multi-agent coordination, Multi-agent systems, Sports science, Training Load Management, Wearables
|
|
