Safety Assessment of Rehabilitation Robots: A Review Identifying Safety Skills and Current Knowledge Gaps

The assessment of rehabilitation robot safety is a vital aspect of the development process, which is often experienced as difficult. There are gaps in best practices and knowledge to ensure safe usage of rehabilitation robots. Currently, safety is commonly assessed by monitoring adverse events occurrence. The aim of this article is to explore how safety of rehabilitation robots can be assessed early in the development phase, before they are used with patients. We are suggesting a uniform approach for safety validation of robots closely interacting with humans, based on safety skills and validation protocols. Safety skills are an abstract representation of the ability of a robot to reduce a specific risk or deal with a specific hazard. They can be implemented in various ways, depending on the application requirements, which enables the use of a single safety skill across a wide range of applications and domains. Safety validation protocols have been developed that correspond to these skills and consider domain-specific conditions. This gives robot users and developers concise testing procedures to prove the mechanical safety of their robotic system, even when the applications are in domains with a lack of standards and best practices such as the healthcare domain. Based on knowledge about adverse events occurring in rehabilitation robot use, we identified multi-directional excessive forces on the soft tissue level and musculoskeletal level as most relevant hazards for rehabilitation robots and related them to four safety skills, providing a concrete starting point for safety assessment of rehabilitation robots. We further identified a number of gaps which need to be addressed in the future to pave the way for more comprehensive guidelines for rehabilitation robot safety assessments. Predominantly, besides new developments of safety by design features, there is a strong need for reliable measurement methods as well as acceptable limit values for human-robot interaction forces both on skin and joint level