NinjaTurtle has ink-blue ears, and with his dark camera eyes mounted on the side of the rounded body, the little robot looks rather cute. The machine helps neuroinformatics and cognitive scientists from Ulm testing special algorithms based on human perception and cognition. The researchers hope that this will make processing of visual and auditory sensor data more robust, faster and more efficient.
“Still, the human brain is one of the most effective data processing systems in existence. Natural nervous systems are highly effective and superior to many technical systems, especially in the evaluation of sensory impressions,” explains professor Heiko Neumann. The deputy head of the Ulm Universities’ Institute of Neuroinformatics is one of the successful applicants, like his colleague Marc Ernst, head of the Department of Applied Cognitive Psychology. With their VA-MORPH project, the scientists want to transfer neurobiological functions of the brain to robotic and information technology systems. The focus is on the development of so-called neuro-morphic algorithms, whose structure and mode of operation are oriented to the human brain and its elementary components, the neurons. Starting point is the question of how visual and auditory sensor streams can be processed, fused and used for technical purposes, for example for spatial orientation and navigation.
“Human perception is not clocked as in technical systems, but works on an event-based basis. Relevant is what changes over a certain period of time. From the enormous stream of information we are confronted with, the brain selects only those who are “relevant to survival” and make sense in the respective situation,” explains Ernst, a cognitive psychologist. This does not happen like with a conventional camera, where the spatial environment is captured by single images. The biological hearing process is just as complex and no less “data-economical”. To hear and process auditory impressions, the brain combines sensory signals with expectations from different contexts of experience and calculates the information into a multi-sensory overall impression. “The integration of these sensory data streams is a masterpiece of the brain. Once we have understood exactly how this works, we can try to transfer these functionalities to technical systems for sensor data processing,” the Ulm researchers summarize their scientific mission.
To generate the neuromorphic algorithms, the scientists have developed biologically plausible learning methods that can be used to filter out the relevant information from all sensory data. Now, the researchers want to find out how practical and efficient the algorithms based on human models are. To this end, these algorithms are implemented on the robot platform and tested on simple orientation tasks. The small ninja bot now has the task of visiting and collecting certain visual and acoustic “landmarks” without being distracted or distracted by background noise and visual obstructions.
In order to implement the algorithms, particular computer architectures are used. With this so-called “brain-inspired hardware”, the processor and memory are not separated, as is the case with conventional computers. Rather, they work together like neurons and their synaptic connections in the brain. This allows the data to be processed much faster and more efficiently. Thanks to cooperation partners, the group has access to scientific equipment that is probably unique for universities. For example, IBM Research Almaden (USA) provides neuro-morphic chip architectures in the field of brain-inspired computing. In addition, the scientists can access a hardware platform from the EU-funded Human Brain Project and, through company IniLabs, to special neuro-morphic sensors.
If all goes well, NinjaTurtle will soon be able to move without being disturbed by noise, even under difficult visibility conditions, and will only consume a fraction of the computing capacity and storage space required by conventional computer architectures.
The project is part of the Neurorobotics programme of the Baden-Württemberg Foundation and is funded with 500,000 euros.