Prior identification method of safety performance boundary for autonomous vehicles based on Safety Threat Field.

Objective: Identifying the safety performance boundary (SPB) of autonomous vehicles (AVs) is crucial for verifying the coverage of scenario-based tests for AVs. Researchers proposed posteriori methods based on collected crash scenarios that demonstrated promising advancements in identifying the SPB. Nevertheless, the search for crash scenarios is often rendered complex and time-consuming due to the "curse of dimensionality and rarity." To address this limitation, this paper introduces the Safety Threat Field-based Prior Identification Method (STF-PIM) to identify the SPB a priori.

Method: Firstly, the STF model is constructed to quantify the safety risks posed by various scenario elements, where background vehicles and other obstacles are considered as sources of safety threats. By defining the Safety Threat Field Potential Energy (STFPE), we establish a state space that maps the ego vehicle's response to a crash in an actual physical scenario as its traversal through the state space to dissipate the STFPE. The maximum STFPE that the ego vehicle can dissipate is defined as its safety capacity. Next, we calibrate the safety capacity of the ego vehicle using critical crash scenarios. Through analysis, we find that in critical crash scenarios, the ego vehicle makes its utmost effort to avoid a crash, resulting in the dissipation of the STFPE from its initial value in the first frame to exactly zero in the last frame. Consequently, the threshold STFPE for all crash scenarios (the minimum STFPE that leads to a crash) can be equated to the safety capacity of the ego vehicle. By comparing the initial STFPE of a scenario with this threshold, crash scenarios can be identified without having to examine every possible scenario in the scenario space.

Results: A cut-in scenario is performed in simulation to validate the proposed STF-Prior Identification Method (STF-PIM) for identifying the SPB a priori. Simulation results show that the proposed STF-PIM successfully describes the SPB of the ego vehicle without traversing the entire scenario space.

Conclusions: The proposed STF-PIM enables the calibration of the safety capacity of the tested ego-vehicle through a small number of critical crash scenarios, thereby providing an a priori description of the SPB.