Fraunhofer: NeurOSmart enables safe human-machine collaboration

The Fraunhofer Institute for Silicon Technology (ISIT), along with other institutes, has jointly developed the NeurOSmart platform, which is intended to enable safe collaborative work between humans and machines. The system consists of Lidar laser sensors, whose signals are evaluated by self-developed, energy-efficient, AI-powered chips. The shared workspace can thus be monitored, and machines can be slowed down or stopped in case of danger.

NeurOSmart is a joint development of the Fraunhofer Institutes for Photonic Microsystems (IPMS), for Electronic Circuits and Systems (ECS), for Machine Tools and Forming Technology (IWU), and for Intelligent Analysis and Information Systems (IAIS). ISIT coordinated the project.

The system monitors the workspace that humans and machines must share when collaborating. To keep the risk to humans as low as possible, the area is monitored from a bird's-eye view by Lidar lasers. The laser emits short pulses in the near-infrared range for this purpose. The laser reflections are used for three-dimensional distance measurement.

To cover the entire workspace with a manageable number of Lidar lasers, the NeurOSmart system uses movable MEMS mirrors (Micro-Electro-Mechanical Systems), whose performance and energy efficiency have been increased by Fraunhofer scientists in the project. For the mirrors, the researchers use piezoelectric aluminum scandium nitride (AIScN) with a layer thickness of one micrometer.

The captured data is used to generate a high-resolution 3D image of the work environment. However, a large amount of data must be processed for this. Before this can happen, the system uses AI-powered algorithms to focus on identifying areas where something is most likely to happen. The sensors can then be aligned to reduce the amount of data to be processed and require less computing power for processing.

Data evaluation using Neuromorphic Computing

The data is then evaluated using self-developed AI chips, whose processors consist of several small computing units connected in a matrix and located directly in the sensor system. The technology of Neuromorphic Computing is based on the functioning of the human brain. Each chip acts as a "thinking cell", according to ISIT. The data evaluation process can thus be very fast. Only a few milliseconds pass from the reception of signals through their evaluation by the AI models developed by Fraunhofer to the control of a machine such as a robot arm. Potentially dangerous robots can thus be slowed down or completely stopped before they collide with a person.

The collaboration between humans and robots with NeurOSmart in industrial environments is intended to be not only safer but also more agile and efficient. This is because the fast and precise detection of hazards only slows down or switches off the robot when it is necessary.

According to the scientists, NeurOSmart could also be used in other application scenarios in the future. The NurOSmart technology is energy-efficient, so it could be used in drones and agricultural sensor systems, for example.

(olb)