Post: A Virtual Rat With an AI Brain Could Mean Better Robots

A digital rat. Image created by Decrypt using AI Science has built a better rat, controlled by an artificial intelligence (AI) brain.

This virtual rodent, developed by researchers at Harvard and Google’s DeepMind AI lab, is capable of accurately mimicking the movement of real rats, a significant achievement that promises to advance our understanding of how brains control complex, coordinated movements.

It’s a feat that even today’s most advanced robots struggle to replicate, and the team believes their findings could bring future robots vastly improved agility.

Professor Bence Ölveczky from Harvard’s Department of Organismic and Evolutionary Biology led the effort, utilizing high-resolution data recorded from real rats to train the artificial neural network. His lab is dedicated to the mechanistic exploration of how the brain makes limbs move.

"We found that neural activity… was better predicted by the virtual rodent’s network activity than by any features of the real rat’s movements, consistent with both regions implementing inverse dynamics," the researchers wrote.

According to the study published in the journal Nature , the virtual rat simulation was crafted using the MuJoCo physics simulator, incorporating realistic forces such as gravity to mimic real-world conditions. The artificial neural network that drives the virtual rat’s movements was trained on inverse dynamics models, enabling it to predict neural activity in real rats with high accuracy.

The report noted that the results can help scientists interpret neural activity across different behaviors and relate it to principles of motor control. The principles derived from studying the virtual rat have implications for the development of improved robotic control systems.

“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,” the research reads. Image: Google Deepmind “We’ve learned a huge amount from the challenge of building embodied agents: AI systems that not only have to think intelligently but also have to translate that thinking into physical action in a complex environment,” Matthew Botvinick from Google Deepmind told the Harvard Gazette. “It seemed plausible that taking this same approach in a neuroscience context might be useful […]

A Virtual Rat With an AI Brain Could Mean Better Robots