To explore the mysteries of how the brain controls movement, scientists from Harvard University and Google's Deep Thinking Laboratory have collaborated to create a "virtual rat" - a biomechanically realistic digital model of a rat. This "rat" has an artificial brain that can move around like a real rodent. This achievement represents a new method for humans to study how the brain operates. The relevant paper was published in the latest issue of the journal Nature. The movements of humans and animals are extremely agile and can be considered a miracle in evolution, and to this day, no robot has been able to deeply imitate them. Scientists believe that humans guide movement through the brain. For example, when a person reaches out for a cup of coffee, the brain quickly calculates the trajectory that their arm should follow and converts it into a motion command. This time, team members worked closely together and trained an artificial neural network using high-resolution data from real rats as the "brain" of virtual rats. This "brain" was used to control the virtual body in a physical simulator called MuJoco. The network is input with the reference trajectory of the desired motion and learns to generate forces, ultimately training an artificial neural network to achieve the so-called inverse dynamic model. This allows "virtual rats" to mimic various behaviors, even those that have not been explicitly trained. The virtual control network can accurately predict the same neural activity in the brain as the real rat. Combined with deep reinforcement learning and 3D motion tracking of free acting animals, this simulation is realized. This simulation research may open an undeveloped field of virtual neuroscience. In this field, AI simulated animals behave like real animals after training, providing a convenient and completely transparent model for studying neural circuits, and even deducing how these circuits are damaged in diseases. Simulation platforms can also serve as an example to help people design better robot control systems. (Lai Xin She)