John Harwell

Dr. Harwell is a researcher specializing in multi-agent systems. His PhD thesis developed new theoretical tools for measuring, modeling, controlling, and (critically) predicting the behavior of bio-inspired multi-agent systems from small (≤ 5 agents) to large (≥ 10, 000 agents across scales, and targeted applications to foraging and construction tasks in dynamic, dangerous, and unknown environments.

His research interests lie in the investigation of behavior in interacting multi-agent systems, at the intersection of task allocation, mathematical modeling, complexity theory, robotics, and swarm intelligence. Broadly:

• Understanding the fundamental principles of large systems, including the “unpredictable” behaviors which emerge as systems interact with their environments in non-trivial ways, to develop better models of collective behavior.
• Developing mathematical models for predictive control of large-scale multi-agent systems from first principles which are robust enough to cross the simulation-reality gap. Of particular interest are applications in dangerous or unstable environments (e.g., mining, space exploration), or dynamic environments (i.e., those which are modified by the system as it operates), such as agriculture, autonomous construction, debris/waste removal.
• Accelerating development of multi-agent systems through better tooling, measurements.

More specifically:

1. Behavioral modeling and analysis: Modeling explicit and implicit information flow in multi-agent systems, information latency. Prediction of collective behavior from first principles on non-trivial problems such as foraging, for both memory-less agents and agents with finite memory (forward problem). Deriving agent controllers to obtain a desired collective behavior (inverse problem).
2. Self-healing systems: Mitigation of malfunctioning sensors/actuators, autonomous convergence recognition. Robust anomaly recognition: task-independent identification of misbehaving individuals in the presence of dynamic workloads. Rebalancing task distributions after handling anomalies or whet tasks change/new tasks are dynamically introduced.
3. Engineering better multi-agent systems: How to create insightful measurements of measurements of system behavior such as self-organization, scalability, convergence, robustness, etc., and how these can be made prescriptive so they can be used in automated design methods.
4. Long-term multi-robot system autonomy: How to increase the ability of these systems to operate for long periods of time without/with minimal human intervention or supervision. Optimal foraging theory, control barrier function approaches, optimization of battery usage under uncertainty.

In addition to theoretical multi-agent systems, Dr. Harwell is also interested in bringing elements of software engineering into research in order to accelerate research progress and reproducibility through automation.

Dr. Harwell's areas of expertise include:

• Swarm intelligence
• Swarm robotics
• Mathematical modeling of multi-agent systems
• Distributed algorithm design and data structures
• Software engineering
• Real-time embedded systems
• Computational optimization

His personal website is here.