Sensory integration and management of uncertainty in robot safety systems: A review

Abstract

Today, as many new application areas for robotic systems emerge, safe operating conditions for human personnel engaged in installation, programming and maintenance of a robot have become a critical consideration. Robots make use of multiple sources of sensory information. However, the diverse and complex sensors of a robot system supply information that is often uncertain, conflicting, or incorrect. Therefore, it is difficult to interpret and aggregate this information into a purposeful and homogenous robot's workspace description. Thus, in the area of robotics, sensory integration and management of uncertainty have become problems of immense practical importance. Furthermore, due to limited computational power and a lack of effective algorithms for processing sensory information, multisensory systems suffer from a relatively low level of real-time performance. To achieve a real-time response in robot systems, it is necessary to consider new technology such as artificial neural networks that enable parallel computing, brain-like control, and representation. This article reviews recent literature on robot safety, focusing on collision avoidance, management of uncertainty in sensory integration, and application of neural networks in the above areas.