Harvard and Google’s DeepMind Create Agile Virtual Rat, Revolutionising Neuroscience and Robotics

Harvard neuroscientists, in collaboration with Google’s DeepMind AI lab, have created a virtual rat with an artificial brain that can move like a real rodent. The team, led by Professor Bence Ölveczky, used high-resolution data from real rats to train an artificial neural network to control the virtual rat in a physics simulator. The research, published in Nature, could provide insights into how the brain controls movement and could be used to engineer better robotic control systems. The project was financially supported by the National Institutes of Health.

The Creation of a Virtual Rat: A New Approach to Studying Brain Control of Movement

In an effort to understand the complex coordination and control of movement in humans and animals, a team of neuroscientists from Harvard University, in collaboration with Google’s DeepMind AI lab, have developed a virtual rat with an artificial brain. This digital model of a rat, which is biomechanically realistic, can move in a manner similar to a real rodent. The team, led by Bence Ölveczky, a professor in the Department of Organismic and Evolutionary Biology, used high-resolution data from real rats to train an artificial neural network, effectively creating the virtual rat’s “brain”. This artificial brain was then used to control the virtual body in a physics simulator called MuJoco, where gravity and other forces are present.

The Promise of the Virtual Rat: Accurate Predictions and New Insights

The results of this innovative approach, published in Nature, are promising. The researchers found that the activations in the virtual control network accurately predicted neural activity measured from the brains of real rats producing the same behaviors. This achievement represents a new approach to studying how the brain controls movement, leveraging advances in deep reinforcement learning and AI, as well as 3D movement-tracking in freely behaving animals. The collaboration with Google’s DeepMind was instrumental in this project, as they had developed a pipeline to train biomechanical agents to move around complex environments.

The Application of Inverse Dynamics Models: Guiding Movement in the Virtual Rat

The team trained the artificial neural network to implement what are called inverse dynamics models, which scientists believe our brains use to guide movement. For instance, when we reach for a cup of coffee, our brain quickly calculates the trajectory our arm should follow and translates this into motor commands. Similarly, based on data from actual rats, the network was fed a reference trajectory of the desired movement and learned to produce the forces to generate it. This allowed the virtual rat to imitate a diverse range of behaviors, even ones it hadn’t been explicitly trained on.

The Potential of Virtual Neuroscience: Studying Neural Circuits and Disease

These simulations may open up a new area of virtual neuroscience, where AI-simulated animals, trained to behave like real ones, provide convenient and fully transparent models for studying neural circuits. This could also extend to studying how such circuits are compromised in disease. While the primary interest of Ölveczky’s lab is in fundamental questions about how the brain works, the platform could also be used to engineer better robotic control systems.

The Future of the Virtual Rat: Autonomy and Advanced Understanding

A potential next step for this research could be to give the virtual rat autonomy to solve tasks similar to those encountered by real rats. This could provide insights into how such tasks are solved and how the learning algorithms that underlie the acquisition of skilled behaviors are implemented. The team hopes to use the virtual rats to test these ideas and help advance our understanding of how real brains generate complex behavior. This research received financial support from the National Institutes of Health.