Safe Autonomous Overtaking Maneuver based on Inter-Vehicular Distance Prediction and Multi-Level Bayesian Decision-Making

Reliable Risk Management for Autonomous Vehicles based on Sequential Bayesian Decision Networks and Dynamic Inter-Vehicular Assessment

“…Section II). In previous work [1], [6], we proposed a probabilistic framework which assesses the overall surrounding environment by evaluating the collision risk with all observed vehicles, plan driving maneuvers considering predictions of road user trajectories and make the decision on the most suitable actions. This sequencing of decisions is handled by the means of a robust Sequential Level Bayesian-Decision Network (SLBDN) (cf.…”

“…Section II-B). Moreover, using the specific properties of the dynamic predicted inter-distance profile (DPIDP) between vehicles (defined in [6], cf. Section II-A.1), used as risk metric input into the safety verification, anomaly detection criteria are defined (cf.…”

“…1). The probabilistic decision process is based on the current situation assessment, using the Extended Time To Collision The second decision is a part of the Safety Verification Decision Level (SVDL) where for each time step T s , while the maneuver execution starts, a safety-checking regarding the action chosen in the MDL and a verification of the coherence of the maneuver with the predicted pre-planned trajectory is performed through an improved definition of anomaly detection criteria based-DPIDP [6] (cf. Section II-A), used to detect and compensate for possible failure of the perceptive module or unexpected behaviors.…”

“…4) Criteria for anomaly detection: Further, contrary to what have been proposed as anomaly criteria in [6], where, at each sample time, is calculated the current available lateral error between the actual inter-distance profile (AIDP) and the DPIDP, it is proposed in this paper to exploit the dynamic of these measures progress to have more pertinent risk assessment and management strategies (cf. Section II-A).…”

“…This is mainly caused by measurement uncertainties and unexpected maneuvers of other traffic participants. For this reason validating the safety of self-driving vehicles while applying safety verification methods can prove the coherence of the vehicles behavior, reduce remaining risks and the need for extensive testing and more importantly allow us to plan evasive maneuver, in real-time [1], [6], [7].…”