Autonomous systems rely on learning from experience to automatically refine their strategy and adapt to their environment, and thereby have huge advantages over traditional hand engineered systems. We use reinforcement learning (RL),  a principled mathematical framework for sequential decision making under uncertainty, to develop intelligent agents capable of acting in very dynamic and unknown environments.

The objective in reinforcement learning is to maximize the agent’s long term expected rewards given the current model of the environment. An autonomous agent continuously updates its model of its environment in the light of data collected while operating. Such flexible, data driven models are able to generalise to new situations, but predictions inevitably contain some level of uncertainty, as they are based on a finite (typically small) sample of data. The agent calculates optimal actions (the policy) based on a reward function and the dynamics model. The construction of the optimal policy depends crucially on a faithful representation of the uncertainties (precision) of the learned dynamical models. Probabilistic (Bayesian) models provide a principled and practical approach and Bayesian decision theory provides an automatic framework for constructing policies.

Modern computational methods in machine learning allow learning of autonomous agents for a wide range of application areas, ranging from robotics, process control, autonomous game agents, smart cities, autonomous vehicles and complex decision support systems. One of the most crucial practical properties of such systems is sample efficiency, that is, the ability to learn as quickly as possible from noisy and limited data.

At we are developing autonomous decision making systems based on recent progress in probabilistic methods in machine learning, game theory  and modern computational tools. The agents we develop are able to interact in dynamic complex environments possibly populated by other agents. Crucially, we recognise that the ultimate goal for autonomous systems are the quality and reliability  of the decisions made, and rational decision making is the driving force requiring data collection, model construction and inference.