Bio-behavioral Team Dynamics Measurement System: Multimodal Sensing, Dynamical Systems Modeling, and Machine Learning Pipelines to Predict and Characterize Team Performance

The DARPA OP TEMPO program seeks to accelerate warfighter readiness by supplying instructors with objective, automatic assessments of team performance during simulation training. To that end, we created the Bio-behavioral Team Dynamics Measurement System (BioTDMS), a multimodal sensing and analytics pipeline that discovers bio-behavioral “signatures” emanating from within the human body and through team-member interactions that predict team performance. BioTDMS employs a layered symbolic dynamics model that converts time-aligned neural, cardio-respiratory, eye tracking, and verbal data, collected using a multimodal sensor suite. Moving-window entropy and mutual information computed across the symbolic sensor space yield real-time metrics that quantify team adaptability following perturbation (e.g., “training injects”) and distribution of team members’ influence across biological and behavioral subsystems. These features feed a multitask, multi-kernel learning engine that refines performance prediction while preserving explainability through team construct mapping and a command-line user interface. We present preliminary results from field testing a full physical and computational implementation of BioTDMS during Fire Support Team (FiST) training exercises at the U.S. Marine Corps Air-Ground Combat Center, Twentynine Palms, CA. An onsite team instrumented five-person FiST crews with multimodal sensor suites. Sensor data were processed by BioTDMS for real-time and post hoc analytics. BioTDMS currently accounts for 90 % of variance in a subjective team performance assessment made by instructors, with improvements expected upon further refinements of BioTDMS modeling components. These findings demonstrate BioTDMS’s potential as an operational tool for automatic, objective team assessments. Future assessments within air combat teams, including configurations with human-autonomy teaming, will evaluate the generalizability of BioTDMS.