Real-time Temporal Logic-based Planning for Multi-agent Autonomous Systems in Partially-known and Uncertain Environments

自动驾驶汽车或无人驾驶飞行器(UAV)等自动驾驶系统被视为解决交通运输(如减少拥堵和排放)以及降低人类在消防和搜救等安全关键任务中的风险的主要问题的解决方案。自2010年以来，自动驾驶汽车行业的投资已超过2000亿美元，预计到2045年将成为数十亿美元的行业[31]，无人机行业也是如此[3]。然而，目前这些系统的成功应用大多局限于简单、孤立的环境，如笼子外的仓库、农村空域和专用交通路线。现实世界，特别是城市环境，由于人类操作系统的原因，仍然不确定，并且不断以部署之前未曾见过的方式发展。在如此复杂的环境中，自动驾驶系统远不能完全自动运行，最近一系列涉及自动驾驶汽车的备受瞩目的事故就是明证，这导致美国国家公路交通安全管理局(NHTSA)在2021年8月启动了一项调查。我提出的研究计划的长期目标是使自主系统能够在这种不确定、不断发展和先验未知的环境中规划它们的运动，同时满足复杂的目标。这些要求可能包括：1)空间(例如，避免设立禁飞区)、2)时间(例如，森林灭火无人机每5分钟监视2个高风险区域)、3)反应性(例如，灭火无人机必须在接下来的2分钟内撤离飞机落水的区域)，4)跨多个机构(例如，在因烟雾而能见度有限的区域，至少2架无人机应确认存在更多火灾)。如何以数学上合理的方式简洁而正确地指定这些目标，然后自动生成自主系统的运动计划，以便在复杂的现实世界环境中实现这些目标，这是一个挑战。该计划旨在通过在使用形式(基于逻辑的)规范语言来捕获此类操作要求以及在满足这些形式规范的运动规划方法的基于优化的方法方面取得根本性进展来解决这些技术挑战。一个特别的重点将是开发高效的方法，这些方法实时在线运行，并允许在具有挑战性的环境中执行这些任务的多个自治系统之间进行协调。虽然加拿大拥有快速增长的机器人产业，但在毕马威2020年自动驾驶汽车准备指数中，加拿大排名第12，在技术和创新方面落后于其他国家。这一研究计划将在现实系统的自主系统决策基础方面取得根本性进展，并将培训学生和研究人员，使其具备在快速增长的自主系统部门领导地位的能力，包括机器人(地面和空中)和汽车行业。

Autonomous systems, such as self-driving cars or unmanned aerial vehicles (UAVs), are seen as the solution for major problems in transportation (e.g, reducing congestion and emissions) and in reducing risk for humans in safety-critical tasks such as firefighting and search-and-rescue. The self-driving car industry has seen over US$200 billion invested in it since 2010, and is expected to be a multi-billion dollar industry by 2045 [31], as is the UAV industry [3]. Currently however, successful applications of these systems are mostly limited to simple, isolated environments such as caged-off warehouses, rural airspace, and dedicated traffic routes. Real-world, especially urban environments remain uncertain due human-operated systems and constantly evolve in ways not seen prior to deployment. In such complex environments, autonomous systems are far from capable of fully autonomous operation, as seen by the recent spate of high-profile accidents involving self-driving cars which led the US National Highway Traffic Safety Administration (NHTSA) to open an investigation in August 2021. The long-term goal of my proposed research program is to enable autonomous systems to plan their motion in such uncertain, evolving, and a priori unknown environments while satisfying complex objectives. These could include requirements that are in: 1) space (e.g., avoiding a no-fly zone), 2) time (e.g., forest-fire-fighting UAVs surveilling 2 high-risk areas every 5 minutes), 3) reactive (e.g., fire-fighting UAVs must clear out of area where an aircraft will drop water on within the next 2 minutes), and 4) across multiple agents (e.g., in areas with limited visibility due to smoke, at least 2 UAVs should confirm the presence of additional fires). It is a challenge to both concisely and correctly specify such objectives in a mathematically sound manner and then to automatically generate motion plans for the autonomous systems to carry out these objectives in complex real-world environments. This program will aim to address these technical challenges by making fundamental advances in the use of formal (logic-based) specification languages for capturing such operating requirements, and in optimization-based methods for motion planning methods that satisfy these formal specifications. A particular focus will be on developing efficient methods that run online in real-time and allow for co-ordination between multiple autonomous systems carrying out these tasks in challenging environments. While Canada has a fast-growing robotics industry, it ranked 12th in KPMG's 2020 Autonomous Vehicle Readiness Index, trailing other countries in technology and innovation. This research program will make fundamental advances in the underpinnings of decision-making for autonomous systems for real-world systems and will train students and researchers to be equipped to lead in fast-growing autonomous system sectors, including robotics (ground and aerial) and the automotive industry.