Neuromorphic Computing

Our research activities center on improving the learning capability and intelligence of technical systems, whether in the manufacturing, automobile and robotics. To carry out these activities, we rely on findings from the field of neurobiology and apply software methods from the area of artificial intelligence (AI) and the subarea of deep learning. We also develop algorithms and software for highly energy-efficient neuromorphic hardware and its use in the area of machine learning.            

1.  Neuromorphic robotics
   Controlling mobile or industrial robotics in an energy efficient way with low latency is critical for tomorrow's autonomous robots. More AI will have to be processed on device, as robots and objects get more intelligent and it will have to save on battery life. Neuromorphically controlled robotics is an essential topic with problems such as SLAM, motion control, online learning, … In this research line, we focus on motion control, with two projects, one about robot swimming (INRC1), one about object insertion with a robotic arm using reinforcement learning (INRC3). On the simulation side, in the frame of the Human Brain Project, we participate in the development of the Neurorobotics Platform that is a central tool for all our projects.
    
2. Neuromorphic vision
   Event-based sensing, and in particular vision, is essential to robotics. So this second research line actually serves the first one, though it is more recent. In this research line, that could be applied in automotive, security, military, smartphone, medicals, household electronics, etc, we focus on mobile robotics (drones in the FAMOUS project), industrial robotics (robotic arm in ELEANOR) and human-machine interaction in an upcoming project. We also explore opportunities in space applications and automotive.

These tutorials are aimed at engineers and R&D managers in large and small industrial companies. They present a revolutionary technology for on-board artificial intelligence (edge-AI) in terms of energy efficiency and latency: neuromorphic computing. Gains of several orders of magnitude are possible!

Moreover, certain neuromorphic chips enable on-chip learning, and therefore on-line learning and system adaptability. For example, a system that has been pre-trained for some people can be adapted for others in just a few seconds online. In this way, systems can self-improve. We are convinced that neuromorphic computing offers solutions for many embedded AI applications in automotive, aerospace and medical technology, robotics, logistics, consumer electronics and, of course, smartphones.

Neuromorphic hardware