Self-Driving Car Perception Platform

点击翻译按钮获取中文摘要

Self-driving cars are looming on the horizon, but there is still much research to be done before they become a reality. Many of the outstanding technical challenges involve imbuing these vehicles with robust perception so that they might (i) know where they are, (ii) recognize dynamic objects, and (iii) understand the state of the world around them. While satellite-based positioning can help with the first task, it is widely accepted that on-board sensors will be necessary to fully solve these perception tasks. This project proposes to acquire a roadworthy research vehicle instrumented with several different types of sensors (cameras, laser rangefinders, radar, inertial measurement units); the vehicle will be used to gather large datasets in order to advance the perceptive capabilities of self-driving cars. While some such datasets exist (e.g., the well-known KITTI dataset prepared by co­applicant Urtasun - downloaded 10,000 times, used by 300 research groups worldwide), this is only a starting point. Toronto enjoys a wide variety of different seasonal conditions (particularly winter), making it a unique test area to push the boundaries of self-driving car perception. Gathering datasets in varying conditions is necessary to push research in this area, and execute the research aims of the applicants and their industrial/government partners. This project represents an important and new collaboration between the Dept. of Computer Science and the Faculty of Applied Science and Engineering and will surely be a high-profile mechanism to showcase the University of Toronto and Canadian robotics research.