Collaborative robots and current human–robot interaction systems, such as exoskeletons and teleoperation, are key technologies with profiles that make them likely security targets. Without sufficient protection, these robotics technologies might become dangerous tools that are capable of causing damage to their environments, increasing defects in work pieces and harming human co-workers. As robotics is a critical component of the current automation drive in many advanced economies, there may be serious economic effects if robot security is not appropriately handled. The development of suitable security for robots, particularly in industrial contexts, is critical.
Collaborative robots, exoskeletons and teleoperation are all examples of robotics technologies that might need close collaboration with humans, and these interactions must be appropriately protected. There is a need to guard against both external hackers (as with many industrial systems) and insider malfeasance. Only authorised users should be able to access robots, and they should use only those services and capabilities they are qualified to access (e.g. those for which they are appropriately cleared and trained). Authentication is therefore a crucial enabling mechanism. Robot interaction will largely be ongoing, so continuous rather than one-time authentication is required.
In robot contexts, continuous biometrics can be used to provide effective and practical authentication of individuals to robots. In particular, the working behaviour of human co-workers as they interact with robots can be used as a means of biometric authentication.
This thesis demonstrates how continuous biometric authentication can be used in three different environments: a direct physical manipulation application, a sensor glove application and a remote access application. We show how information acquired from the collaborative robot's internal sensors, wearable sensors (similar to those found in an exoskeleton), and teleoperated robot control and programming can be harnessed to provide appropriate authentication. Thus, all authentication uses data that are collected or generated as part of the co-worker simply going about their work. No additional action is needed. For manufacturing environments, this lack of intrusiveness is an important feature.
The results presented in this thesis show that our approaches can discriminate appropriately between users. We believe that our machine learning-based approaches can provide reasonable and practical solutions for continually authenticating users to robots in many environments, particularly in manufacturing contexts.
