A robotic dog walking under the control of human neurons grown in a lab is no longer just science fiction. The Australian start-up Cortical Labs used its CL1 biological computer to steer the movements of a four-legged robot, transferring to a physical body what the system had learned in a simulation.
The CL1 is a box holding living cells wired to a silicon chip. In earlier tests it was trained to head towards a target in a simulated two-dimensional environment. It is the same basic demonstration with which the company had already shown, in 2021, that its neurons could play Pong. This time the training data was used to govern a real robot, a step that moves the technology from the game to the physical world.
The most interesting part concerns learning. Ferrante Neri, of the Education University of Hong Kong, who was not involved in the research, noted in the company's paper on arXiv, not yet peer-reviewed, experiments in which the neural network improved over time even when fed the same digital data. According to Neri, this suggests it is being successfully trained for the task. He cautions, however, that these are promising results under limited experimental conditions: handling the complexity of real-world navigation remains a distant goal.
The reason the company keeps pursuing this path is energy use. Cortical Labs argues that a biological computer could require less power than a digital alternative, a decisive advantage in robotics, where battery life is often the main limit. The same argument was highlighted at the biological data centre in Singapore, where CL1 units draw about 30 watts each, less than a calculator.
The project also connects to growing interest in computing that mirrors the brain: just as AI models such as ChatGPT are inspired by biological neural networks, here real cells are used to replicate their digital counterparts. The path remains experimental, and the results must still pass the scrutiny of the scientific community.
No comments yet. Be the first!