Formal Verification-based Risk Assessment for Industrial Human-Robot Collaboration

Human-robot collaboration (HRC) imposes potential frequent physical interaction and/or close proximity between the two agents. Most likely, and specifically for unstructured environments, changing layouts and dynamic task allocation, the prediction of hazardous conditions may be difficult or incomplete. Nonetheless, conducting a thorough risk assessment on the mechanical hazards—physical harms to the human operator caused by the robot—is essential for collaborative systems, to define preventive or responsive mitigation mechanisms within the system. In previous works [1], [2], we have defined a methodology, SAFER-HRC, that applies formal verification for hazard identification and risk analysis of contact hazards. SAFER-HRC creates formal models of HRC applications via the TRIO temporal logic [3] and uses an automated verification tool, called Zot [4], to exhaustively search their state space for hazardous situations. Given a UML model of the application based on a specific profile notation [5], SAFER-HRC translates it to a logic model containing: (i) logic formulae that describe a discrete representation of operator, robot and the layout, the most important entities of collaborative systems, and the executing job; (ii) formulae modeling the significant hazardous situations as described in ISO 10218-2 and ISO/TS 15066; (iii) formulae modeling human error phenotypes [6]; (iv) a formal replication of the ISO/TR 14121-2 risk estimation procedure; (v) formulae describing risk reduction measures (RRM) for collaborative modes as described in ISO/TS 15066. SAFER-HRC does not replace human risk assessors, but it provides an automated assistant that helps them detect hazardous situations and compute the overall risk of the system. Figure 1 shows a walk-through of SAFER-HRC where, starting from UML diagrams, a formal model is automatically generated and verified. In case the model requires additional RRMs, manual intervention is needed by the human assessor to choose the best-suited RRMs for each situation