Dynamics and geometry of locomotor control and learning
Heike Stein
Institut des Systèmes Intelligents et de Robotique (ISIR) – Paris

Abstract
Locomotion on natural terrain requires animals to navigate irregular surfaces with sparse footholds. Yet, most theoretical and experimental work has focused on locomotion on flat surfaces. We address two questions: how can locomotor theory be generalized to complex terrain, and what constitutes locomotor learning in this setting?
To study locomotor learning, we longitudinally tracked paw trajectories of mice walking on a motorized ladder. We model inter-limb coordination with a system of switching weakly coupled oscillators, revealing that locomotion on complex terrain is governed by multiple coexisting dynamical regimes rather than a single stable attractor. Geometrically, gait on complex terrain lives on a high-dimensional manifold, rather than the low-dimensional manifold of regular locomotion on flat terrain. Within this framework, different dynamical regimes allow for more or less stable coordination dynamics, and mice gradually abandon unstable regimes in favor of slow, persistent coordination dynamics. We show that this is driven by an optimization of single-limb swing timing, which drives transitions and favors slow dynamics. We propose that locomotor learning on complex terrain is requires navigating the geometry of an existing high-dimensional behavioral manifold toward gait patterns that maximize stability.

Invited by Lorenzo Fontolan & David Robbe
Monday 14 September 2026 at 11.00 am – Inmed conference room