08. March 2018

Robots with Brains

Artificial Intelligence

fortiss has developed a new simulator for the "Human Brain Project", enabling neuroscientists to "transplant" their brain models into virtual robots. The unique feature of these virtual robots is that they are equipped with a brain that works with pulsed neural networks (known as SNN, or "Spiking Neural Networks"). These are even closer to their biological model than the first generation neural networks known from machine-based learning. The latter uses databases to identify and learn patterns, whilst SNN promises better coding of nerve impulses and therefore a finer coordination of movements. The idea is that if robots move in a similar way to humans, they will be easier to build and safer to control in the future.

The main objective of the European Commission's Human Brain Project (HBP) is to decipher and simulate the human brain. Technological tools will also be developed to study and better understand the brain as well as the thinking and behaviour of humans and animals. The knowledge gained will benefit the fields of biology, medicine and IT: scientists in these disciplines will be provided with new tools to explore novel therapies for brain diseases and to develop new computer and robotic technologies.

Information and communication technologies play a crucial role in the HBP, which is split into different sub-sections. As part of the neurorobotics sub-project, a hardware and software infrastructure is being developed for scientists to connect brain models with robots and test them in virtual worlds. The project envisages a future where robots use sensors to perceive and process information in the same way as their biological models.

Spiking Neural Networks

As part of the project, neuroscientists have developed a brain model of a mouse with 75 million neurons. fortiss computer scientists have integrated a simplified version of it into their robot simulator. The special feature of this new simulator is that the robot brain works with spiking neural networks. These simulate the behaviour of biological nerve cells, which transmit their signals as short electrical impulses from cell to cell - comparable to a digital code. This enables neuroscientists to study the activity of neurons while the robot moves in a virtual world.

"The new simulator gives us the opportunity to investigate the complex processing of movements in the brain. We can apply the knowledge gained from it to future robot or computer architectures," explains fortiss group leader Axel von Arnim.

Numerous applications are conceivable. Obvious areas of implementation are in human-machine interaction and in the field of prosthetics. "The ability to simulate the interface between the body and a prosthesis in future is a major development for us. Thanks to the pulsed neural networks, the prosthesis and nerves speak the same language. This makes it possible to optimise artificial limbs faster than ever before," explains the computer expert.

Video at
https://www.youtube.com/watch?v=rvw9Siblrw4&list=PLFfa5EHopIFosLhZa3HxGQzo1JyM-MhUq&index=3


Background:

fortiss engineers are playing a leading role in developing the neurorobotics platform. The cooperating partners are the Ecole Polytechnique Fédérale de Lausanne (EPFL), the Technical University of Munich, the Scuola Superiore Sant'Anna (SSSUP) in Pisa and the FZI Research Centre for Information Technology in Karlsruhe.

The EU-funded HBP began in 2013 and is split into 12 sub-projects. The project consortium consists of over 110 international research institutions and partners from the industry. The budget is approx. EUR 1 billion. The end date is 2023.

Contact

Online & Marketing:
Kathrin Kahle

PR & Communication:
Dr. Evdoxia Tsakiridou