A safety monitoring concept for fully automated driving

Abstract

Safe motion planning for automated vehicles requires that a collision-free trajectory can be guaranteed. For that purpose, we propose a monitoring concept that would ensure safe vehicle states. Determining these safe states, however, is usually a computationally demanding task. To alleviate the computational demand, we investigate the possibility to compute the safe sets offline. To achieve this, we leverage backward reachability theory and compute the N-step robust backward reachable set offline. Based on the current disturbances, we demonstrate the possibility to adapt this set online. The safety guarantees are then provided by computing the robust one-step forward prediction of the state vector and checking if these states are members of the adapted safe set. The numerical example demonstrates that the approach is capable of avoiding hazardous vehicle states under an unsafe motion planning algorithm.

Formal methods to comply with rules of the road in autonomous driving: State of the art and grand challenges

“…Safety specifications Monitoring for compliance with safety specifications is presented in [17,63,64,134]. The authors of [17] use backward reachability analysis to construct the monitor.…”

“…Safety specifications Monitoring for compliance with safety specifications is presented in [17,63,64,134]. The authors of [17] use backward reachability analysis to construct the monitor. Monitoring for Multi-Lane Spatial Logic is considered in [134], where the authors show that formula satisfaction can be mapped to feasibility of formulas in the first-order theory of real-closed fields.…”

“…We believe that these tables makes it easy for the reader to find a paper using a specific technique in a particular application area. For example, if she wants to find a paper that uses online monitoring from STL specifications with applications to lane keeping, then she would use Table 4 to find the papers that apply online monitoring to lane keeping: [17] and [18]. She would then check which of these papers appear at the intersection of the "STL" column and the "lane keeping / changing" row in Table 2.…”

Collision Avoidance of Autonomous Driving at Low Speed in the Near Field of Vehicle

“…The safety of the following path needs to check whether the path is collision-free. A common approach is reachability analysis [6] [7]. The reachability analysis focuses mainly on the control problem of finding all of the reachable sets for a given initial state.…”

Safety Monitoring of Neural Networks Using Unsupervised Feature Learning and Novelty Estimation

“…For tasks such as control and planning this may involve monitoring the vehicle states in order to reject trajectories during high levels of input sensor noise resulting in unsafe motion planning, see e.g. [16] and [13]. Similarly monitoring methods have been developed for deep learning approaches rejecting predictions when considered unreliable due to the environmental conditions.…”