Intervention Request Planning with Operator Capability Model for Human-Automation Cooperative Recognition.

Human-automation cooperation in the recognition phase of the autonomous driving system (cooperative recognition) has been proposed to address the challenges in the conventional cooperation method, e.g., taking over vehicle control. In cooperative recognition, the operator intervenes in the recognition of obstacles and risks that are difficult for the automated system alone. To realize cooperation and maximize the performance of the overall cooperative system, human tasks must be carefully allocated taking into account human processing capability and state in addition to driving safety, efficiency, and comfort. Since the human states are not directly observable, we formulate this problem as a Partially Observable Markov Decision Process (POMDP). Through simulator experiments, we showed that designing reward functions in the POMDP model that are biased operator decisions leads to inappropriate intervention requests, and we presented a solution. Furthermore, the intervention request scheduled by the POMDP model was able to reduce the intervention request time while maintaining driving comfort compared to the myopic policy, which requests intervention from the closest target, and also the POMDP model could adapt to the operator’s state.