Animals have exquisite control over their bodies, allowing them to perform a wide range of behaviors. However, the way in which this control is implemented by the brain remains unclear. Advancing our understanding requires models that can relate principles of control to the structure of neuronal activity in well-behaved animals. To facilitate this, we constructed a “virtual rodent,” in which an artificial neural network operates a biomechanically realistic model of the rat in a physical simulator. We used deep reinforcement learning to train the virtual agent to imitate the behavior of freely moving rats, allowing us to compare the neural activity recorded in real rats to the network activity of a virtual rodent imitating their behaviour. We found that neuronal activity in the sensorimotor striatum and motor cortex was better predicted by the network activity of the virtual rodent than by the movement characteristics of the real rat, consistent with both regions highlighting works an opposite dynamic. Furthermore, latent network variability predicted the structure of neuronal variability across behaviors and provided robustness consistent with the minimal intervention principle of optimal feedback control. These results demonstrate how physical simulation of biomechanically realistic virtual animals can help interpret the structure of neural activity across behavior and relate it to theoretical principles of motor control.
Here is the new Nature article by Diego Aldardo, et.al. Via @sebkrier.
