Minds for Mobile Agents (M4MA)

The Minds for Mobile Agents project uses tools from cognitive decision modelling and behavioural data science to understand how diverse groups of pedestrians move around realistic spatial and social environments to fulfill complex goals.

The project’s “predictive pedestrian” model (GitHub for project members, for access email Charlotte Tanis) endows simulated pedestrians with a first-order theory-of-mind, basing their step choices on predictions about the behaviour of other agents. The agent’s “minds” can be flexibly specified using an easily extensible utility maximising framework with parameters that have clear psychological interpretations, making the behaviour of the model explainable.

In the animation below basic functions such as attempting to maintain a preferred speed and interpersonal distance, being attracted in the direction of a goal and away from directions blocked by other agents, and social attractions, such as following others going in the same direction, enable simulated pedestrians to autonomously navigate around a supermarket to obtain items on a shopping list.

Agents moving through a supermarket
M4MA example: Customers moving through a supermarket

Each pedestrian has dynamic route planning abilities, enabling them to strategically navigate to a set of goal locations. The animation below shows route planning for a series of shopping goals (red dots) via way points (+ symbols) while respecting one-way rules in some aisles.

Agent model setup
Model setup: each pedestrian dynamically plans their route to navigate to goal locations

We are now calibrating the model with data from experiments providing high temporal and spatial resolution measurements of the behaviour of real pedestrians performing different spatially tasks. Bayesian hierarchical modelling will be used to determine the distributions of model parameters over people in order to account for individual differences.

M4MA was originally developed to understand and test social-distancing protocols with the Smart Distance Lab, and is now being applied more broadly to problems involving lower-density pedestrian flows where individual differences play a more crucial role than in the highly crowded scenarios with simple tasks addressed by most pedestrian models.  


M4MA Team

The research is led by Associate Professor Dora Matzke and Révész Visiting Professor Andrew Heathcote, with an IXA proof-of-concept grant supporting PhD student Charlotte Tanis. Further team members include PostDoc Researcher Tessa Blanken, Associate Professor Michael Lees, and Professor Denny Borsboom.