Development of a Semi-Autonomous Robotic System to Assist Children with Autism in Developing Visual Perspective Taking Skills

Robot-assisted therapy has been successfully used to help children with Autism Spectrum Condition (ASC) develop their social skills, but very often with the robot being fully controlled remotely by an adult operator. Although this method is reliable and allows the operator to conduct a therapy session in a customised child-centred manner, it increases the cognitive workload on the human operator since it requires them to divide their attention between the robot and the child to ensure that the robot is responding appropriately to the child's behaviour. In addition, a remote-controlled robot is not aware of the information regarding the interaction with children (e.g., body gesture and head pose, proximity etc) and consequently it does not have the ability to shape live HRIs. Further to this, a remote-controlled robot typically does not have the capacity to record this information and additional effort is required to analyse the interaction data. For these reasons, using a remote-controlled robot in robot-assisted therapy may be unsustainable for long-term interactions. To lighten the cognitive burden on the human operator and to provide a consistent therapeutic experience, it is essential to create some degrees of autonomy and enable the robot to perform some autonomous behaviours during interactions with children. Our previous research with the Kaspar robot either implemented a fully autonomous scenario involving pairs of children, which then lacked the often important input of the supervising adult, or, in most of our research, has used a remote control in the hand of the adult or the children to operate the robot. Alternatively, this paper provides an overview of the design and implementation of a robotic system called Sense- Think-Act which converts the remote-controlled scenarios of our humanoid robot into a semi-autonomous social agent with the capacity to play games autonomously (under human supervision) with children in the real-world school settings. The developed system has been implemented on the humanoid robot Kaspar and evaluated in a trial with four children with ASC at a local specialist secondary school in the UK where the data of 11 Child-Robot Interactions (CRIs) was collected. The results from this trial demonstrated that the system was successful in providing the robot with appropriate control signals to operate in a semi-autonomous manner without any latency, which supports autonomous CRIs, suggesting that the proposed architecture appears to have promising potential in supporting CRIs for real-world applications.