Reachset Conformance of Forward Dynamic Models for the Formal Analysis of Robots

Abstract

Model-based design of robotic systems has many advantages, among them faster development cycles and reduced costs due to early detections of design flaws. Approximate models are sufficient for many classical robotic applications; however, they no longer suffice for safety-critical applications. For instance, a dangerous situation which has not been detected by model-based testing might occur in a human-robot co-existence scenario since models do not exactly replicate behaviors of real systems-this problem arises no matter how accurate a model is, since even disturbances and sensor noise can cause a mismatch. We address this issue by adding non-determinism to robotic models and by computing the whole set of possible behaviors using reachability analysis. By using reachset conformance, we automatically adjust the required non-determinism so that all recorded behaviors are captured. For the first time this approach is demonstrated for a real robot.