Where did I park? Reinforcement Learning approaches to flexible spatial navigation.

We are delighted that the following research, undertaken in collaboration with the University of Nottingham School of Psychology, has recently been accepted for publication in the British Neuroscience Association Journal: Brain and Neuroscience Advances!

Both humans and non-human animals are flexible in navigating space, necessitated by the need to continually find food and shelter. This includes the ability to return to specific locations based on as few as one single experience. But how is it that we remember, for example, where we left our car in a large crowded car park?

University of Nottingham PhD student Charline Tessereau, along with her supervisors Associate Professor Reuben O’Dea, Professor Stephen Coombes (School of Mathematical Sciences) and Associate Professor Tobias Bast (School of Psychology), are exploring the computational principles involved in a Reinforcement Learning (RL) framework of spatial navigation. RL (a subfield of Artificial Intelligence and Neurosciences) algorithms can support aspects of the neural computations that are involved in the rapid learning undertaken in spatial navigation.

Charline comments that “mathematical modelling is a perfect complement to behavioural, lesion and recording studies as it allows us to test how systems can produce similar behaviours, giving us insights on possible computations underlying behaviours in the brain.”

“RL agents learn to navigate through predicting spatial navigation outcomes. My algorithm uses a simultaneous learning of where the goal is and how to get there. Our paper investigates the biological realism of every part of the network.”

Understanding the mechanisms involved in spatial navigation has a wide range of applications from preventing disorientation of Alzheimer’s patients to producing a navigating robot. The next step of this ongoing research at the University of Nottingham is to get closer to behaviours through making predictions that can be tested on animal models in collaboration with the School of Psychology and Neuroscience.