KnowGo: An Adaptive Learning-Based Multi-model Framework for Dynamic Automotive Risk Assessment

Abstract

In autonomous driving systems, the level of monitoring and control expected from the vehicle and the driver change in accordance with the level of automation, creating a dynamic risk environment where risks change according to the level of automation. Moreover, the input data and their essential features for a given risk model can also be inconsistent, heterogeneous, and volatile. Therefore, risk assessment systems must adapt to changes in the automation level and input data content to ensure that both the risk criteria and weighting reflect the actual system state, which can change at any time. This paper introduces KnowGo, a learning-based dynamic risk assessment framework that provides a risk prediction architecture that can be dynamically reconfigured in terms of risk criterion, risk model selection, and weighting in response to dynamic changes in the operational environment. We validated the KnowGo framework with five types of risk scoring models implemented using data-driven and rule-based methods.