Co-design of Reachability Analysis and Trajectory Planning for Collision Avoidance Systems

道路车辆的防撞系统有可能接管完全控制，必须面临许多挑战。其中包括对环境的不确定测量，其他交通参与者的不确定未来运动，以及安全运动的解决方案空间通常很小。小的解决方案空间是故意的，因为防撞系统应该只在驾驶员几乎没有可能将车辆带入安全状态时才使用。虽然运动规划可以被认为是相当好的研究，但在紧急情况下情况完全不同：最先进的运动规划器的计算时间越长，解空间越小。这与在关键情况下需要小的计算时间相矛盾，因此通常找不到安全的解决方案，并且崩溃是不可避免的。相反，可达性分析变得越快，解空间越小（可达性分析返回动力系统的可能解的集合）。在这个项目中，我们开发了一种新的可达性分析和运动规划的协同设计，以实现在危险情况下具有小计算时间的运动规划器。通过使用可达集，可以更好地修剪基于图的规划器的搜索空间，并更好地指导规划器使用基于梯度的连续优化。我们还将使用可达集来识别狭窄的通道，以确保运动规划器在不引起任何碰撞的情况下通过这些通道。此外，我们自动推导出车辆可以无限期停留而不会造成碰撞的安全状态。这使得可以为无限时间范围提供安全的运动计划。为了进一步节省计算时间，我们的目标是修复不安全的运动计划，即，仅更改关键部件，以便仅需要对修复的部件重新运行碰撞检查。所提出的概念将通过自动生成的关键情况（及其演变）进行集中测试。自动测试生成也将在服务器上实现，以便其他研究人员也可以测试他们的运动规划器。这将首次有可能对其他方法进行基准测试，因为目前还没有针对自动驾驶道路车辆的标准化测试。所获得的结果也可用于自动驾驶，以确保在关键情况下的安全解决方案。此外，其他新的智能系统也必须保证安全运行，例如部分自动化医疗机器人，在生产中实现安全人机共存的系统，以及智能电网，都受益于我们的研究结果。

Collision avoidance systems for road vehicles potentially taking over the full control have to face many challenges. Among them are uncertain measurements of the environment, uncertain future movements of other traffic participants, and the often small solution space for a safe motion. The small solution space is intentional since collision avoidance systems should only engage when a driver has almost no possibility left to bring the vehicle into a safe state. While motion planning can be considered as rather well researched, the situation is completely different in emergency situations: The computation time of state-of-the-art motion planners is the larger, the smaller the solution space is. This contradicts the need of small computation times in critical situations, thus safe solutions are often not found and a crash is inevitable. In contrast, reachability analysis becomes the faster, the smaller the solution space is (reachability analysis returns the set of possible solutions for a dynamical system). In this project, we develop a novel co-design of reachability analysis and motion planning to realize a motion planner with small computation times in dangerous situations. By using reachable sets, one can better prune the search space of graph-based planners and better guide planners using gradient-based continuous optimization. We will also identify narrow passages using reachable sets to make sure that the motion planner passes those without causing any collisions. Further, we automatically derive safe states in which a vehicle can stay indefinitely without causing a collision. This makes it possible to provide safe motion plans for infinite time horizons. To further save computation time, we aim at repairing unsafe motion plans, i.e., only change critical parts so that only collision checks are required to be re-run for the repaired part.The proposed concept will be intensively tested by automatically-generated, critical situations (and their evolvement). The automatic test generation will also be implemented on a server so that other researchers can test their motion planners as well. This would make it possible for the first time to benchmark other approaches since no standardized tests yet exist for automated road vehicles. The obtained results can also be used for automated driving to guarantee safe solutions in critical situations. Also other new intelligent systems, which also have to guarantee a safe operation, such as partially automated medical robots, systems realizing safe human-robot co-existence in production, as well as smart grids, benefit from our results.