Embodied Agents for NeuroAI
The nervous system evolved to control complex bodies in dynamic environments. However, many studies of motor control isolate components rather than model the full brain-body-environment loop. Understanding natural behaviour requires integrated models linking neural control, biomechanics, and real-world dynamics. Advances in 3D motion capture, artificial neural networks, and physics simulation offer a method to bridge this gap. In this talk, I present my work on training embodied rodent agents in MuJoCo to emulate naturalistic behaviour and, furthermore, apply those learned representations towards downstream task solving.