Co-design of Reachability Analysis and Trajectory Planning for Collision Avoidance Systems
Final Report Abstract
Motion planning approaches for automated road vehicles have to face many challenges. Besides the uncertain measurements of the environment and uncertain future movements of other traffic participants, these systems face the problem that the computation time increases when the situation becomes more dangerous. This is due to the reduced solution space, making it difficult for classical motion planners to find a feasible solution. However, especially in dangerous situations, a short computation time would be desirable. We solved this problem by combining reachability analysis with motion planning (reachability analysis returns the set of possible solutions for a dynamical system). In contrast to classical motion planners, reachability analysis becomes the faster, the smaller the solution space is. We developed a novel codesign of reachability analysis and motion planning to realize a motion planner with small computation times in dangerous situations. By using reachable sets, we can better prune the search space of graph-based planners and better guide planners using gradient-based continuous optimization. To further improve the safety of automated road vehicles, we also developed a method to 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 researched novel methods to repair unsafe motion plans, i.e., only change critical parts so that only collision checks are required to be re-run for the repaired part. To the best of our knowledge, we provided the first collision-free motion planner whose computation time decreases in critical situations – as opposed to all other motion planners, which would fail in such situations. Obviously, the critical situations are those that are especially crucial so that we believe that our findings are significant and of societal value. Our novel findings were published in eight high-profile papers, out of which three are published in renowned journals. In addition, we organized the first international competition on motion planning for autonomous vehicles in realistic scenarios.
Publications
-
“Computing the Drivable Area of Autonomous Road Vehicles in Dynamic Road Scenes.” In: IEEE Transactions on Intelligent Transportation Systems 19.6 (2018), pp. 1855–1866
S. Söntges and M. Althoff
-
“Falsification-Based Robust Adversarial Reinforcement Learning.” In: Proc. of the 19th International Conference on Machine Learning and Applications. 2020, pp. 205–212
X. Wang, S. Nair, and M. Althoff
-
“Safe Reinforcement Learning for Autonomous Lane Changing Using Set-Based Prediction.” In: Proc. of the 23rd IEEE International Conference on Intelligent Transportation Systems. 2020, pp. 1–7
H. Krasowski, X. Wang, and M. Althoff
-
“Using online verification to prevent autonomous vehicles from causing accidents.” In: Nature Machine Intelligence 2 (2020), pp. 518–528
C. Pek, S. Manzinger, M. Koschi, and M. Althoff
-
“CommonRoad-RL: A Configurable Reinforcement Learning Environment for Motion Planning of Autonomous Vehicles.” In: Proc. of the IEEE International Intelligent Transportation Systems Conference. 2021, pp. 466–472
X. Wang, H. Krasowski, and M. Althoff
-
“Sampling-Based Optimal Trajectory Generation for Autonomous Vehicles Using Reachable Sets.” In: Proc. of the IEEE International Intelligent Transportation Systems Conference. 2021, pp. 828–835
G. Würsching and M. Althoff
-
“Sampling-Based Trajectory Repairing for Autonomous Vehicles.” In: Proc. of the IEEE International Intelligent Transportation Systems Conference. 2021, pp. 572–579
Y. Lin, S. Maierhofer, and M. Althoff
-
“Using Reachable Sets for Trajectory Planning of Automated Vehicles.” In: IEEE Transactions on Intelligent Vehicles 6.2 (2021), pp. 232–248
S. Manzinger, C. Pek, and M. Althoff
