Lane-Change Regulation via Reinforcement Learning for Platoon Formation in Mixed Freeway Traffic

Journal Article (2026)
Author(s)

Anastasios Bithas (TU Delft - Civil Engineering & Geosciences)

Emmanouil Kampitakis (National Technical University of Athens)

Eleni I. Vlahogianni (National Technical University of Athens)

Research Group
Transport, Mobility and Logistics
DOI related publication
https://doi.org/10.1109/TITS.2026.3728593 Final published version
More Info
expand_more
Publication Year
2026
Language
English
Research Group
Transport, Mobility and Logistics
Journal title
IEEE Transactions on Intelligent Transportation Systems
Page Views
18
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

State of the art freeway lane management strategies that allocate Connected Autonomous Vehicles (CAVs) into designated lanes enable platoon formation and increase capacity. However, the lane changes required during strategy initialization can induce capacity drops and upstream congestion. To mitigate this problem, this study proposes two Reinforcement Learning enabled policies that regulate the lane changes of CAVs during the strategy initialization phase: a. a homogeneous restriction policy, which uses an RL-agent to directly determine a global lane-change probability for eligible CAVs, and b. a disruption-aware restriction policy, which combines a global lane-change probability, determined by an RL agent, and a metric that predicts the disruption of a lane change, determined by ML methods, to calculate the individual lane-change probability. The proposed approaches are evaluated using microscopic traffic simulation, under various traffic demands and CAV market penetration rates. The results show an increase in the average platoon length in the leftmost lane by up to 22.2% compared to no application of lane management methods, while averting the capacity drop of unrestricted lane management. The efficiency of the proposed methods is consistent across various CAV market penetration rates, highlighting the flexibility and generalization of agents trained at a baseline penetration rate.

Files

– Personal use only – Dutch Copyright Act (Article 25fa)
warning

File under embargo until 16-03-2027