Machine Learning for the next generation of paediatric wheelchairs

Independent mobility is one primary signifier impacting multiple health outcomes in children. Beyond core mobility-related health outcomes, the acquisition of gross motor milestones and independent mobility is important in a child's ongoing emotional, psychosocial and cognitive development. However, development of paediatric assistive mobility is hindered by three main factors: i) Lack of research in paediatric assistive area (mobility sector in specific) in comparison to adult services; ii) Lack of holistic, convergent and innovative thinking within paediatric mobility services. Iii) Slower pace of assistive market in adopting new and emerging technologies and design principles compared to other areas. As a result, paediatric mobility is still a largely unexplored research area, with available products not taking full advantage of modern design, manufacturing, control and software technologies. Powered wheelchairs are the assistive mobility equipment with maximum potential for improving multiple health outcomes in children and offering unlimited independent mobility.
This project aims to explore how sensing and machine learning can contribute to the creation of the next generation of paediatric powered wheelchairs. Two intertwined research questions motivate the project: i) what sensor and data are needed to enable active assistance in paediatric powered wheelchairs? ii) what sensors and data will enable a deeper understanding of the children's learning process so that better guidance can be provided to future users?
The answer to the first question will involve mechatronic design for the choice and installation of the sensors, and machine learning to use the sensor data to implement functionality such as obstacle avoidance, assisted driving and safety features (e.g. emergency stop in presence of dangers). The aim is to lower the barrier of adoption of powered wheelchairs, making them accessible to children at a younger age and/or having a broader spectrum of disabilities.
The answer to the second question will involve again mechatronic design of the sensor system and machine learning to map inputs such as obstacles around a child to outputs represented by the choice of action of the child driving the wheelchair. Understanding this mapping, and its evolution as a child gets better at driving the wheelchair, will contribute to a better understanding of the users learning process that, in turn, will enable the creation of better training guidelines for the future users of these devices.