Declarative Rule-based Safety for Robotic Perception Systems

Mobile robots are used across many domains from personal care to agriculture. Working in dynamic open-ended environments puts high constraints on the robot perception system, which is critical for the safety of the system as a whole. To achieve the required safety levels the perception system needs to be certified, but no specific standards exist for computer vision systems, and the concept of safe vision systems remains largely unexplored. In this paper we present a novel domain-specific language that allows the programmer to express image quality detection rules for enforcing safety constraints. The language allows developers to increase trustworthiness in the robot perception system, which we argue would increase compliance with safety standards. We demonstrate the usage of the language to improve reliability in a perception pipeline and evaluate it against manually written rules on embedded hardware. The language allows the vision expert to concisely express the safety-related constraints and thereby bridging the gap between domain experts and certification authorities.