The RONIN project put its innovative robot-assisted training approach to the test in real clinical settings. Over several weeks, the robot worked with a total of 85 participants diagnosed with Autism Spectrum Disorder (ASD) on a set of activities divided in three training protocols: planning self-care routines (such as preparing an orange juice, getting ready for school), navigating social situations (such as interaction with another passenger on public transport), and facing potentially stressful contexts (such as oral exams at school). In the first two protocols, the robot worked alongside participants as a partner, helping them break down everyday self-care or social situations into manageable action steps. Together, they built sequences of actions by combining action steps represented as images glued onto foam cubes that were exchanges between the robot and the child. This activity of sequencing action steps to form an entire daily activity helped participants practice planning abilities. In the social scenario, the robot also introduced a bit of challenge by mixing in distractions or even small “mistakes” to see if participants could identify and correct them, thus encouraging them to think about social rules, such as recognizing when someone might need help. The robot’s calm, consistent voice and steady movements made it easier for participants to focus and stay engaged, reducing the frustration that can come with learning new skills. In the third, most advanced protocol, namely the simulation of an exam, the robot took on the role of a supportive examiner. The robot asked questions based on specially prepared comics, that children studied at home, adapting its approach depending on how the participant responded. This setup helped participants practice staying calm under pressure while working through challenges, all in a supportive environment where mistakes were part of the learning process. Across all training protocols, the robot’s interactions were designed to feel like a collaboration, giving the chance to participants to build confidence and improve social and cognitive skills to use in real-world challenges of everyday life.
To test the effectiveness of the training scenarios, participants’ progress was compared to a control group that received standard therapy alone. Participants who trained with the robot showed significant improvements in their ability to plan, organize, and verbally describe goal-directed action sequences. The training also delivered promising results in improving participants' emotional well-being. Overall, the project demonstrated that a humanoid robot can be a powerful ally in helping adolescents with developmental disabilities build confidence and essential skills to live more independently.
Beyond the clinical results, RONIN has also technical achievements. All training scenarios were designed to leverage iCub’s grasping, handover, and speech capabilities, all controlled by an human operator. To this aim, a full-stack application consisting of back-end and front-end has been developed. The back-end interfaces with the robotic platform to implement the robot behaviours and makes them available to the front-end where they can be controlled by the operator using a custom graphical user interface (GUI). The GUI allowed operators to control iCub’s movements and speech dynamically, facilitating adaptive responses to user requests and enhancing the realism of the training scenarios. Additionally, in the oral exam simulation protocol, a Large Language Model (LLM) was integrated into the GUI to help the operator quickly generate responses. This ensured the robot could provide instant hints and answers, keeping the training scenario realistic and engaging. Both the back-end and the front-end are part of the PyiCub framework and have been released under open-source permissive licenses to be used and extended by the robotics community.
