William Warren (Brown University), “The Dynamics of Perception and Action: From Pedestrian Interactions to Collective Behavior”

It’s a perplexing time in the study of visual perception. On the one hand, there is a resurgence of models that freely posit a priori structure in the visual system, including priors, generative world models, and physics engines. On the other hand, there is the astonishing a posteriori success of deep neural networks trained only on natural images and image sequences. Although their performance offers an existence proof of the sufficiency of image information for certain visual tasks, the black box of deep learning does not easily offer up that information or how it’s extracted by the visual system.

A science of perception depends on understanding the visual information that is available in natural environments and is used to guide natural behavior. I propose that we take seriously James Gibson’s information hypothesis: For every perceivable property of the environment, however subtle, there must be a variable of information, however complex, that uniquely specifies it. The project is to identify the information that the visual system uses to perceive and act within the constraints of a species’ ecological niche.

Two decades ago I decided to work out a test case to see whether an information-based account of a natural behavior could be sustained. In this talk I will offer a status report on our effort to build a model of visually controlled human locomotion – a pedestrian model – that scales up from individual behaviors like steering and obstacle avoidance, to pedestrian interactions like following and collision avoidance, to the collective behavior of human crowds. Surprisingly, linear combinations of these nonlinear components can account for the emergence of more complex behavior, such as self-organized ‘flocking’, crowd bifurcations, and stripe formation in crossing flows.

Bio

Bill (he/him) earned his undergraduate degree at Hampshire College (1976), his Ph.D. in Experimental Psychology from the University of Connecticut (1982), did post-doctoral work at the University of Edinburgh, and has been a professor at Brown ever since. He served as Chair of the Department of Cognitive and Linguistic Sciences from 2002-10. Warren is the recipient of a Fulbright Research Fellowship, an NIH Research Career Development Award, and Brown’s Elizabeth Leduc Teaching Award for Excellence in the Life Sciences. Warren’s research focuses on the visual control of action – in particular, human locomotion and navigation. He seeks to explain how this behavior is adaptively regulated by multi-sensory information, within a dynamical systems framework. Using virtual reality techniques, his research team investigates problems such as the visual control of steering, obstacle avoidance, wayfinding, pedestrian interactions, and the collective behavior of crowds. Experiments in the Virtual Environment Navigation Lab (VENLab) enable his group to manipulate what participants see as they walk through a virtual landscape, and to measure and model their behavior. The aim of this research is to understand how adaptive behavior emerges from the dynamic interaction between an organism and its environment. He believes the answers will not be found only in the brain, but will strongly depend on the physical and informational regularities that the brain exploits. This work contributes to basic knowledge that is needed to understand visual-motor disorders in humans, and to develop mobile robots that can operate in novel environments. For more information, visit his faculty profile or the VENLab website.

For those who are not in Berlin but would like to join virtually:
https://tu-berlin.zoom-x.de/j/69207754612?pwd=IKxoTdY3dQWccHpce2nA0IsNkNxPHu.1

This talk is part of Aravind Battaje‘s course “Mind, Body, Environment: An Interactive Seminar on Embodied Intelligence,” a seminar introducing to key theories and research highlighting this shift in perspective through invited lectures from experts in the field and interactive sessions.

Photo created with DALL-E by Maria Ott.