Autonomous vehicles need to interact with surrounding vehicles in decision-making, and an important factor to consider is the uncertainty of intention. Situations could be even challenging when predicted intentions exhibits approximate equiprobability, which could lead to either conservative or unsafe decision results. This paper proposes an active interaction approach for decision-making in which the behavior of the ego vehicle is designed to further improve confidence in intention estimations. Firstly, an intention estimator considering active interaction is proposed based on Bayes filter to achieve safe and efficient decision-making under indistinguishable intention probabilities. Then, the estimated intention is utilized to guide the motion prediction of surrounding vehicles, which is integrated into the decision-making model based on Partially Observable Markov Decision Process (POMDP). The lateral and longitudinal actions are generated considering the proposed active interaction scheme, in which behavior of the ego vehicle could actively reduce intention uncertainty. Finally, the proposed method is validated in lane change scenarios in CARLA. The results indicate that the proposed method can interact with vehicles with different driving styles and make safe and efficient decisions.
Active Interaction in Driving: An Intention-Aware Decision-Making for Autonomous Vehicles
2024-09-24
2308868 byte
Conference paper
Electronic Resource
English