Safe and efficient robot autonomy in unstructured and dynamic environments

点击翻译按钮获取中文摘要

The potential of autonomous mobile robots is enormous, particularly if they can operate safely and reliably in unstructured environments such as workplaces, public spaces, and roadways. They can be used to increase productivity, provide assistance to those in need, perform repetitive and dull jobs, and to fill the labour shortages seen across industries ranging from manufacturing to agriculture, to trucking. This potential has resulted in a surge of interest in industry, with many of the biggest technology companies such as Amazon, Apple, Google, and Shopify all pursuing robotics and autonomy. However, while robot autonomy is continuously improving, it is still prone to unexpected failures, resulting in unsafe and/or inefficient operation. Fundamental research is needed into how robots plan their motion in uncertain environments, how they assess safety around moving objects and obstacles, how they coordinate their efforts to work on a large-scale, and how they interact with human supervisors to improve performance and ensure safety. The long-term goal driving this research is to achieve safe and efficient operation of multiple robots in unstructured and dynamic environments. This research program will pursue two directions to improve robot autonomy. The first is to develop fundamental advances in robot motion planning around moving obstacles that enable robots to better assess safety and risk. This includes methods to predict risk of future motions, and to reduce that risk by actively sensing critical portions of the environment and by inferring information about the environment through other's actions. The second direction is for robots to utilize human supervisors to improve performance over time. Supervisors are commonly used to (tele-) operate a robot when needed or to provide guidance to a robot using an interface. However, the current paradigm is to use supervisors in a reactive manner, filling in for the deficiencies of the autonomous robots when necessary. This research program will instead consider the problem of improving performance and safety of the collective supervisor-robot team. We will develop algorithms for robots that proactively leverage supervisors when needed, avoid overloading supervisors, and improve performance over time by learning from the supervisor's expertise. The research program will generate new algorithms for autonomy that increase the capability of robots in unstructured environments. These fundamental advances will enable robots to operate safely and efficiently around humans, forming the basis for a range of new robotic applications in industrial cleaning, warehousing and logistics, agriculture, and autonomous driving. The research program will train personnel that are positioned at the forefront of a growing and transformative industry of strategic importance to Canada.