In a remarkable collaboration between Harvard University and Google DeepMind, researchers have developed an artificial neural network that can control a virtual rat’s movements in an ultra-realistic physics simulation. This groundbreaking model mimics how biological brains coordinate complex behaviors and provides unprecedented insights into motor control, cognition, and neurological disorders. The project involved creating an anatomically accurate biomechanical model of a rat in a sophisticated physics simulator called MuJoCo, based on high-resolution motion data from real rats. The artificial neural network, trained using deep reinforcement learning and inverse dynamics modeling, learned to control the virtual rat’s biomechanics, replicating diverse movements observed in biological data. The neural network demonstrated broad generalization capabilities, producing realistic behaviors it was not explicitly trained on. The virtual brain’s neural activity patterns closely matched those of real rat brains, suggesting that it discovered internal models and motor control principles similar to biological brains. This virtual rat brain model allows researchers to probe and perturb an accessible model of the entire brain-body-environment control loop, providing a new paradigm for investigating motor control and brain function. The insights gained could revolutionize neuroscience, robotics, and our understanding of biological intelligence. Additionally, the project highlights the potential of combining advanced machine learning techniques with high-fidelity simulations to tackle complex scientific problems, as seen in related work on nuclear fusion simulations using the TORAX simulator. This approach paves the way for transformative progress in various fields, including materials science, aerospace engineering, and fundamental physics.

AI Revolution
Not Applicable
July 7, 2024
PT10M58S