A Modeled Approach for Online Adversarial Test of Operational Vehicle Safety (extended version)

The scenario-based testing of operational vehicle safety presents a set of\nprincipal other vehicle (POV) trajectories that seek to force the subject\nvehicle (SV) into a certain safety-critical situation. Current scenarios are\nmostly (i) statistics-driven: inspired by human driver crash data, (ii)\ndeterministic: POV trajectories are pre-determined and are independent of SV\nresponses, and (iii) overly simplified: defined over a finite set of actions\nperformed at the abstracted motion planning level. Such scenario-based testing\n(i) lacks severity guarantees, (ii) has predefined maneuvers making it easy for\nan SV with intelligent driving policies to game the test, and (iii) is\ninefficient in producing safety-critical instances with limited and expensive\ntesting effort. We propose a model-driven online feedback control policy for\nmultiple POVs which propagates efficient adversarial trajectories while\nrespecting traffic rules and other concerns formulated as an admissible\nstate-action space. The approach is formulated in an anchor-template hierarchy\nstructure, with the template model planning inducing a theoretical SV capturing\nguarantee under standard assumptions. The planned adversarial trajectory is\nthen tracked by a lower-level controller applied to the full-system or the\nanchor model. The effectiveness of the methodology is illustrated through\nvarious simulated examples with the SV controlled by either parameterized\nself-driving policies or human drivers.\n