Predictive risk estimation for intelligent ADAS functions

The estimation of risk is a central cornerstone in the evaluation of traffic scene situations for intelligent ADAS. This applies to all levels of functions ranging from simple advices and warning functions right down to the evaluation of possible behavior alternatives, planning and autonomous driving. Current risk handling includes probabilistic collision measures, behavior planning using spatial and temporal occupancy grids, and time metrics such as Time-To-X (e.g. TTC, Time-To-Collision). Especially TTX measures are often used to quantify the threat of a future event. In this paper, we develop a theory of the probability of critical future events such as collisions. We show that by introducing the concept of a “survival probability”, we can derive previously heuristic risk measures like TTX, generalize them to temporally continuous risk indicator functions and apply them to several different situations such as e.g. collision risk, risk of passing nearby without collision, risk at intersection entrance or risk in narrow curves. We show how the TTC-based risk can be described as a special case of a general distance-based risk, with the Time-To-Closest-Encounter (TTCE) and the Distance-of-Closest-Encounter (DCE) as modulating parameters. Furthermore, the introduced risk measures can be used for planning under general conditions, e.g. using “risk maps” to evaluate behavior options according to their risk.