We describe a general pattern for high performant level 2 ADAS features, which enables a fast and cheap verification when expanding the feature in small increments. The pattern says that for higher levels of performance of level (1 and) 2, the ADAS feature shall fully control the driving task it is responsible for, without expecting the driver to intervene to compensate for some circumstances. This is any how from the diver’s perspective a minor step, as the high-performant ADAS in almost all situations today successfully and safely fulfils the expected driving task. Still, from a safety argumentation aspect, this small difference is essential, for two main reasons. One major issue is the addressing of over trust. The proposed pattern implements a safety argumentation in accordance with expected level of trust from the responsible human driver. Todays’ existing solutions that compensate for some aspects, but leaving out a few, may cause that the human driver is expected to intervene to compensate rather seldom. From the perspective of any single driver, this need for intervention is so infrequent that over trust is likely to happen, but from a fleet-level perspective not so infrequent that the ADAS can be considered to safely perform the driving task. Another major issue is that this pattern enables an incremental generation of a safety case, with a limited extra effort for a limited delta of an ODD expansion. The main reason for this is that there is no disruptive effect because of the human driver’s capability to compensate. But it is also the case that there are no specifically ODD-hard-coded concepts in the argumentation strategy. The general argumentation strategy looks the same for each iteration of the ADAS function. This means that what is required in each delta from a safety perspective, is the evidence for deltas related to the ODD expansion of the verification task for the safety-related requirements. In this extended abstract, there is a focus on the basic concepts and the resulting effects, and how they go beyond the state of practice. An extensive explanation, together with elaborated examples, how these results are achieved, is given in the full paper.
Sivencrona et al. (2026) studied this question.