University of Waterloo

Published purpose

The ability of an autonomous system to usefully adapt and learn online is a desirable goal for any autonomous vehicle project, especially in terms of the AI-enabled autonomous transportation of goods in obstacle-rich environments. Furthermore, any vision for autonomy project needs decision-making systems to make use of that visual information. This Project will research, develop and demonstrate autonomous vehicle-based algorithms and decision-making systems that could be deployed on autonomous vehicles such as unmanned aircraft systems. While there are autonomous decision-making and reinforcement learning algorithms under active development in the research community, there are two reasons for undertaking this specific Project. First, no one has ever applied the new Legendre Memory Units (LMU) temporal representation system to this task, and the LMU has been shown to produce orders of magnitude of improvement over traditional methods. Second, the computation system resulting from these methods could likely be efficiently implemented in modern hardware, including GPUs, TPUs, and neuromorphic computation.