Developing and Validating a Supervised Autonomous Robotic System for Autism Spectrum Disorders Therapy

Regarding this developing technology, most RAT studies are limited to the WoZ technique in which robots are remotely controlled—unbeknownst to the child—by a human operator [Figure 1 (a)][2],[7]. The WoZ technique allows human therapists to achieve a high level of social interaction without having to use a complex robotic system. However, it requires a significant amount of human workload and not suitable in the long term [8]. It is necessary to increase the level of robot autonomy in RAT research to lessen the human workload and deliver consistent therapeutic experiences [2],[3]. Full autonomy [Figure 1 (b)] indicates that the robot makes decisions and adapts its actions to any situation by itself. This is not feasible at this point because the robot’s actions must be compliant with the therapeutic goals, interaction context, and state of the child, while its action policies cannot be perfect. Furthermore, fully autonomous robotic systems can raise some critical ethical concerns and are not socially accepted by the general public in the context of interacting with children [9],[10]. However, a “supervised autonomy” in which the robot works independently toward achieving given therapeutic goals under a supervisor’s guidance, is achievable, as shown in Figure 1 (c). When necessary, the supervisor can override the robot’s actions before execution to ensure that only therapeutically valid actions are executed.