Development and Application of a Collision Avoidance Capability Metric 2020-01-1207
This paper describes the development and application of a newly developed metric for evaluating and quantifying the capability of a vehicle/controller (e.g., Automated Vehicle or human driver) to avoid collisions in nearly any potential scenario, including those involving multiple potential collision partners and roadside objects. At its core, this Collision Avoidance Capability (CAC) metric assesses the vehicle’s ability to avoid potential collisions at any point in time. It can also be evaluated at discrete points, or over time intervals. In addition, the CAC methodology potentially provides a real-time indication of courses of action that could be taken to avoid collisions.
The CAC calculation evaluates all possible courses of action within a vehicle’s performance limitations, including combinations of braking, accelerating and steering. Graphically, it uses the concept of a “friction ellipse”, which is commonly used in tire modeling and vehicle dynamics as a way of considering the interaction of braking and turning forces generated at the tire contact patches. When this concept is applied to the whole vehicle, and the actual or estimated maximum lateral and longitudinal accelerations of which the vehicle is capable are normalized, the ellipse becomes a circle that represents the boundaries of vehicle performance that can be utilized for driving, including evasive action. When a potential conflict with another object (e.g., another vehicle or pedestrian) is present, the CAC classifies operating areas within the circle as either successful (avoiding a collision) or not successful (resulting in a collision). The capability of a vehicle to avoid a collision is reflected in CAC, as CAC is larger when the range of possible successful avoidance maneuvers is larger and smaller when the range of possible successful avoidance maneuvers is smaller.
Development and derivation of the CAC are described, and various simulated and real-world test scenarios are described and evaluated.
Jordan Silberling, Paul Wells, Atul Acharya, Joseph Kelly, John Lenkeit
Dynamic Research Inc., AAA Northern California Nevada & Utah