Collective decision making of networks of robots

Intelligent machines are everywhere: smart electricity meters at home, and learning algorithms for traffic control, just to name two. Increasingly, these machines make their own decisions, evoking both excitement and concern. The underlying algorithms are often based on our experience with traditional fully controlled robots and industrial plants. This makes autonomous decision-making systems vulnerable to unexpected perturbations, with potentially disastrous consequences. Consider, for example, industrial mass production failures or cascading power-grid blackouts. The need to control the decision-making dynamics of intelligent machines is pressing and demands revolutionary solutions.

This project aims to identify the role of network structures in optimizing multi-robot decisions and to quantify the trade-offs among stability, accuracy, adaptability, and speed; it also plans to enable learning through evolution, e.g. optimization using adaptation. The main techniques include adaptive control and reinforcement learning.

Phd: Rory Gavin