Autonomous vehicles (AV) are rapidly maturing from a demonstration novelty to a commercial product. However, in their race to market, most companies are focused on developing AV for benign environments. There is a major gap in the performance of today's self-driving cars when operating in winter, and thus a major need to accelerate their all-weather capabilities to ensure Ontarians can reap the benefit of this transformative technology.
This research program will focus on the overlooked challenges of advanced perception, prediction, planning, and control in snow-laden environments and during heavy precipitation, and will fundmentally advance robotic robustness in extreme environments. By partnering with GM, LGE, Applanix, and Algolux, the team will accelerate the adoption of autonomous driving technology in Ontario and in all harsh climates around the world.
Theme 1: Sensor Filtering for Object Detection
This theme will develop visual, lidar, and radar pre-processing methods to improve the full perception pipeline performance in adverse conditions. It will also develop novel uncertainty aware object detection methods that continue to perform as conditions degrade and report a reliable assessment of the quality of their output in all operating conditions.
Theme 2: Sensor Fusion for Localization and Tracking
This theme will develop general-purpose, data-driven solutions for egomotion estimation, mapping, localization, and tracking problems that work with a variety of sensors including radars, cameras, and lidars, and will provide perception integrigty estimates for uncertainty aware planning.
Theme 3: Prediction, Planning, and Control
This theme will leverage the uncertainty-aware models of road participants and conditions to predict the behavious of all road participants, and develop progressice and reactive planning and control strategies for autonomous driving throughout the range of adverse weather conditions.