Japanese researchers have created a robot that looks like a brain. It was grown in a lab to help it ‘think like we’.

Experiments at the University of Tokyo revealed that the compact robotic vehicle with wheels was small enough to fit in the palm of a person and was placed in an easy maze. 

The robot was connected with a culture of brain neurons also known as nerve cells that were grown from living cell cultures.  

When these artificial neurons were electrically stimulated, the machine successfully reached its goal – a black circular box.  

A robot was placed on a flat surface with obstacles and was directed toward the goal (bottom right). It steered its way around walls and obstacles - using neurons grown in the lab

A robot was placed on an obstacle-filled surface and directed toward the goal (bottom left). It used neurons from the lab to steer it around walls and obstacles.


A neuron (also known as nerve cell) is an electrically excitable cells that receives, processes, and transmits information via chemical and electrical signals. 

It is approximately a tenth as large as a human hair and one of its basic components. They carry stimuli so that a person can react to the environment.

The stimulation, such as the burning of the finger at the candle flame, is carried by the ascending neuron to the central nervous and then the descending neurons stimulates the arm to remove the finger.  

The robot could also be redirected by an electric impulse if it veered off-track.

The experiments, detailed in a new paper published in Applied Physics Letters, mark a big step forward in the bid to teach intelligence to robots, according to researchers. 

It is the first time that intelligence has been “taught” to a robot by using lab-grown neurons, which are made from living cells. 

The authors write that they have developed a closed-loop method to generate a coherent signal out of a spontaneously active living neuronal cell culture and then embodied the culture using a mobile vehicle robot. 

‘When the robot collided against obstacles or when its goal wasn’t within 90 degrees, an electrical stimulation was applied to the culture. 

“The robot could reach its goal in four different areas.”  

Artificial neurons made from living cells were used to provide a physical reservoir for the robot’s ability to make decisions.

During the trial, the robot was fed homeostatic signals to effectively tell it that everything was going to plan and it was making progress towards the goal. 

If the vehicle veered in the wrong direction or faced the wrong way, neurons in the cell culture were disturbed by an electric impulse

An electric impulse could be sent to neurons in the cell culture if the vehicle turns in the wrong direction. 


Researchers at University of Cambridge have developed “mini brains” which allow them to study a fatal, untreatable neurological disorder that causes paralysis or dementia. 

Amyotrophic lateral sclerosis (ALS), a form of motor neurone diseases (MND), can affect younger individuals. It usually occurs after the age of 40 to 45. 

Scientists can now see how the condition develops well before symptoms start to appear. They can also screen for potential drug candidates.

Continue reading: MND patients have lab-grown’mini brains.

However, if the bot encountered an obstacle, this homeostasis was disrupted with disturbance signals, making the robot awkwardly judder and recalibrate. 

The robot was continuously fed the homeostatic signals, which were interrupted by the disturbance signals, until it successfully completed the maze task. 

The robot couldn’t see the environment or get other sensory information so it was dependent entirely on the electrical trial and error impulses. 

The researchers have showed intelligent task-solving abilities can be produced by ‘physical reservoir computers’ – a physical body (not necessarily a human) that performs computations based on brain signals 

Hirokazu Takahashi is an associate professor of mechanoinformatics at University of Tokyo. 

Artificial intelligence machines that think like we could be created by advances in physical reservoir computing. 

“A brain of [an]Professor Takahashi said that elementary school children are unable to solve math problems in college admission exams. This could be because their “physical reservoir computer” or the brain dynamics is not rich enough. 

A neuron, also known as nerve cell, is an electrically excitable cell that takes up, processes and transmits information through electrical and chemical signals (pictured, an artist's rendering of neurons in the head)

A neuron, also known by nerve cell, is an electrically excitable, or cell that processes and transmits information via chemical and electrical signals. (pictured, an artist’s rendering of neurons in a head)

“Task-solving abilities are determined by how rich a set of spatiotemporal patterns a network can generate.” 

The team believes that physical reservoir computing in this context will help to better understand the brain’s mechanisms, and may lead to the creation of a neuromorphic machine. 

A neuromorphic computing device can imitate the neuro-biological architectures of the nervous system.      


Neuromorphic computing draws its inspiration from our current understandings of the brain’s architecture, and how it makes decisions.

The brain’s neural networks relay information using pulses or spikes. They also strengthen frequent connections and store the changes locally at synapse-interconnections.

Intelligent behaviours emerge from cooperation and competition between multiple regions within the brain’s neural networks and its environment.

This means that brain cells cannot function in isolation. They are not independent of the individual logic gates of processors. These gates can make millions of computations but they work in isolation. 

Recent advances in machine learning have been largely due to improvements in raw power to process binary zeroes and ones.

Loihi is a Neuromorphic computing design by US technology company Intel. It uses an older, yet unproven technology that weights information based on its value.

The firm believes that this “neuromorphic” approach could allow computers of the future to solve problems autonomously through learned experiences.