Lab-Grown Neurons Decode Braille and Steer Robot Dogs
Biohybrid systems combining living brain cells and silicon hardware show new paths for sensory processing and robotic locomotion.
By Letizia Giugliano
2 min readHuman-Edited, AI-Assisted
Recent breakthroughs in biohybrid robotics have demonstrated that cultured human neural tissue can process tactile information and steer physical machines. Researchers from the University of Bristol and Melbourne-based Cortical Labs are pairing biological neurons with silicon electronics to achieve adaptable learning using minimal computational power.
Decoding Braille with human forebrain organoids
In a study published in Frontiers in Artificial Intelligence, researchers at the University of Bristol examined whether tiny, lab-grown forebrain models known as organoids could translate tactile signals, according to New Atlas. These millimeter-wide tissue arrangements consist of approximately 10,000 nerves and supporting cells positioned atop microelectrode arrays. When a tactile sensor scanned Braille characters, the measurements were translated into electrical pulses delivered directly to the neural tissue. By applying a machine learning algorithm to interpret the resulting cellular activity, the setup correctly identified letters with 61% accuracy on a single organoid. Accuracy reached 83% when signals from three organoids were combined, demonstrating an enhanced resilience against introduced electronic noise.
“Robots that are becoming remarkably capable machines at their physical control still struggle with the higher-order control required to learn quickly or adapt when conditions change.”
— Brett Kagan, co-founder and CEO of Cortical Labs
Embodied neurocomputation in simulated environments
While brain organoids processed static tactile inputs, Melbourne biotechnology firm Cortical Labs and Honda Research Institute Europe tackled dynamic learning, as reported by New Atlas. Cultured neurons placed into CL1 biological computers were interfaced with a virtual agent navigating a simulated environment guided by odor gradients. Because biological cells respond with dynamic variability rather than predictable digital precision, the researchers developed the Embodied Neurocomputation Framework to optimize input signals. After analyzing roughly 1,300 electrical configurations across more than 4,000 hours of real-time interaction, 12 signal patterns enabled the biological networks to consistently learn navigation tasks and outperform standard reinforcement-learning algorithms.
Directing quadruped robots through neural firing
Cortical Labs also translated cellular activity into physical control by using firing rates from biological neural networks to guide a robot dog inside a designated space, according to New Atlas. While modern autonomous hardware handles physical locomotion effectively, real-time behavioral decision-making remains a significant computational hurdle for traditional artificial intelligence. Biological neural plasticity enables cells to rewire connections and learn from environmental feedback, offering an energy-efficient alternative to power-hungry computer clusters. The findings on the Embodied Neurocomputation Framework will be presented at the NeurIPS conference in Sydney, Australia.
Read also on The Patent: WiBotic Brings Patented Wireless Opportunity Charging to Autonomous Fleets
