AI Recommendation Systems for Lane-Changing Using Adherence-Aware Reinforcement Learning

TL;DR

This work tackles lane-changing decisions in semi-autonomous driving under partial driver adherence. It introduces an adherence-aware Deep Q-Network that operates within an MDP and online estimates the compliance level $θ ∈ [0,1]$, updating Q-values with a mixture of compliant and baseline actions. Implemented and evaluated in CARLA, the method shows improvements in travel efficiency and total reward over baseline and standard RL under partial adherence. The framework provides a practical, real-time approach to safer, more efficient AI-assisted driving in mixed-traffic environments.

Abstract

In this paper, we present an adherence-aware reinforcement learning (RL) approach aimed at seeking optimal lane-changing recommendations within a semi-autonomous driving environment to enhance a single vehicle's travel efficiency. The problem is framed within a Markov decision process setting and is addressed through an adherence-aware deep Q network, which takes into account the partial compliance of human drivers with the recommended actions. This approach is evaluated within CARLA's driving environment under realistic scenarios.

Figures (5)

• Figure 1: Process of decision-making.
• Figure 2: Software structure.
• Figure 3: Simulation scenario where partial adherence leads to slower traffic, with different types of paths.
• Figure 4: Training loss and smoothed cumulated reward with shaded fluctuation.
• Figure 5: Lane changing decisions.