An IMM-based POMDP decision algorithm using collision-risk function in mandatory lane change

Abstract

In order to make safe and reasonable decisions in some high-risk environments such as the mandatory lane change, we propose an IMM-based partially observable Markov decision process (POMDP) decision algorithm using the collision-risk function which combines the time-to-collision (TTC), the intervehicular time (IT), and the collision function for mandatory lane change. The newly proposed collision-risk function contains two parts: the vehicle impact factor and the collision function, which is used to assess the risk and determines whether the autonomous vehicle collides with surrounding vehicles. The IMM-base POMDP is used for decision-making and we apply the Monte Carlo Tree Search (MCTS) to solve the problem. In the decision-making process, the belief state is obtained by the Interacting Multiple Model (IMM) algorithm. With the collision-risk function and the probability distribution of the states of surrounding vehicles in the future, the proposed POMDP decision algorithm can determine whether the autonomous vehicle accelerates lane changing or decelerates lane changing, and obtain the acceleration corresponding to each path point. Finally, in order to verify the effectiveness of the algorithm, we perform a driver-in-the-loop simulation through Prescan. We use aggressive driver and conservative driver to control the rear vehicle of the target lane, respectively. Simulation results show that the proposed algorithm can accurately predict the accelerations of surrounding vehicles and make safe and reasonable decisions under two scenarios, which is superior to the general POMDP.

Behavior Planning for Autonomous Driving: Methodologies, Applications, and Future Orientation

“…A POMDP based on Interacting Multiple Model (IMM) is proposed for decisionmaking in lane change scenarios [36]. The algorithm considers the risk of collision by using a function based on time to collision (TTC) and Intravehicular Time (IT).…”

How Do Autonomous Vehicles Decide?

“…Huang et al [ 58 ] developed an IMM-based POMDP decision approach for a mandatory lane change maneuver leveraging collision-risk function incorporating the time-to-collision, inter-vehicular time, and the collision function. The suggested collision-risk function was composed of two components: the vehicle impact factor and the collision function, which calculated the likelihood that the AV would collide with nearby vehicles.…”