United States Department of Transportation
i
Lane Changing of Autonomous Vehicles in Mixed Traffic Environments: A Reinforcement Learning Approach
-
2022-02-01
-
[PDF-1.55 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report, 3/1/2021-2/28/2022
-
Corporate Publisher:
-
Abstract:The purpose of this proposal is to develop innovative reinforcement learning control methods for lane changing of connected and autonomous vehicles (CAVs) in mixed traffic. In the proposed framework, before the CAV changes to the target lane, it needs to predict most likely behavior of surrounding vehicles related to the lane change and then determine the optimal path to the other lane. The lane changing maneuver will be completed by solving a linear quadratic regulator (LQR) problem that yields an optimal controller to monitor the CAV’s longitudinal and lateral movements during the lane-changing maneuver. A novel aspect of this research is to reduce the trajectory planning and tracking problem down to the minimization of a cost function that depends on the target way-point in the target lane the CAV will reach. In the proposal, the research team will integrate reinforcement learning and adaptive/approximate dynamic programming methods to solve this data-driven LQR control problem under constraints, without assuming the exact knowledge of surrounding vehicles, while avoiding the curses of dimensionality and modeling of conventional dynamic programming. Thanks to the systematic use of systems and control-theoretic methods, the proposed framework aims to yield desirable lane-changing controllers with guaranteed stability for CAVs from small samples of historical and real-time data.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
