Brain Processing Unit
- The Future Where Biology and Computer Integrate -
Work 2

Experimental Study on Autonomous Robot Control Using Living Neural Networks

This project explores the potential for intelligent control through biological systems, rather than relying on artificial intelligence. The adaptive learning capabilities of living organisms allow them to flexibly respond to dynamic and unpredictable environments, such as real-world settings. By incorporating actual brain neural circuits as the core of a robotic control system, this project aims to establish a novel control architecture.

The biological brain is an exceptional system that continuously processes external information in real time while acting, generating appropriate subsequent behaviors. This project investigates the unique mechanisms of learning and adaptation inherent in living systems, utilizing cerebral organoids—small-scale neural cell networks.

  • 2025
  • Cerebral organoids, quadrupedal robot, camera system, microelectrode array system
  • 3.5m × 3.5m

Description

This exhibition presents an experiment on autonomous robot control using cultured neural cells (cerebral organoids). The system employs cerebral organoids as the central control unit for a quadrupedal robot, enabling it to execute obstacle avoidance behaviors.

A ceiling-mounted camera tracks the real-time position of the robot, converting this data into electrical signals that are transmitted to the cerebral organoid. The learning mechanism functions by associating electrical stimulation with the presence of free space in the robot’s virtual field of view. When an obstacle is detected, the stimulation decreases, and just before a collision, it nearly disappears.

Through this mechanism, the presence of electrical stimulation serves as a "reward," while its reduction acts as a "punishment," allowing the neural cells to autonomously develop obstacle avoidance behaviors.

In the exhibition space, you can witness the live interaction between the autonomous robot and the biological neural network, observing its adaptive learning process in real time. The system gives you the opportunity to directly experience how living neurons process information and dynamically adjust behaviors.

Technical details

Hardware

  • Web camera
  • GPU workstation
    • Robot position analysis
    • Stimulation pattern conversion
    • Visualization
    • Control signal conversion
  • Server (API)
    • Stimulation command generation
    • Pre-processing
    • Spike count
  • High-density microelectrode array system
  • Cerebral organoids (two connected)
  • Projector
  • Unitree Go2

System configuration diagram

Screen capture