Lead: Dr Shane Windsor
Uncrewed air vehicles (UAVs), such as quadcopters, are one of the most visible examples of the future autonomous systems that may soon be operating in public environments. These systems offer enormous potential for tasks such as infrastructure monitoring, surveillance, emergency response, small payload delivery and even personal transport. However, the autonomy and decision-making abilities of these systems is currently limited and well below the level required to operate fully autonomously in real world conditions; particularly those as complex and uncertain as urban environments. However, recent advances in fields such as machine learning offer the potential to increase the autonomy of these systems by allowing them to learn from experience. The combination of aerial robotics and learning based algorithms for flight control creates a series of technical, social and safety challenges:
- Learning based algorithms are trained with a finite set of data and may have unpredictable functionality when encountering situations outside of their training. The development of suitable methods for dealing with these situations in the context of aerial robotics is required
- The urban environment is highly dynamic and unpredictable, yet we need to develop suitable specifications for assessing the functionality of UAVs to ensure they are safe, reliable, and resilient
- When a UAV can learn in operation and evolve in functionality this raises ethical and legal issues in relation to responsibility and liability
At the TAS Functionality Node we are developing a series of flight control systems for small UAVs that incorporate a spectrum of machine learning components. In doing this we are also exploring different design methods for making systems Trustworthy-by-Design.