Multi-Agent Decision Making

Large-scale systems appear in many areas of science and engineering with a plethora of applications. One of the most representative examples is case of robotic multi-agent teams and swarms. Performing optimization at scale is a challenging problem and requires a creative approaches that relies on distributed optimization architectures, parallel programming, and flexible optimal control algorithms for deterministic and stochastic systems. In this research direction we take algorithms such as Differential Dynamic Programming, MPPI and Deep Forward Backward Stochastic Differential Equations and aim to scale them to multi-agent robotic systems at extreme scales and for different types of vehicles, tasks and missions.

Large scale dynamic optimization will always be an open problem as scales will only increase and new computational paradigms will be required to support scalable computation for extreme large-scale decision-making problems in robotics and autonomy. Some highlights of our research include our work on decentralised optimal control algorithms and their scaling to large scale multi-agent systems.

Our most recent work Distributed Differential Dynamic Programming for Mulit-Vehicle Control. In this work, DDP is scaled up to millions of optimization variables. Arxiv version is coming soon!

Best Response MPPI:

Distributed Model Predictive Path Integral Control:

Evolving Cost Functions for Model Predictive Control of Multi-Agent UAVs.