An Enhanced Graph Representation for Machine Learning Based Automatic Intersection Management

Abstract

The improvement of traffic efficiency at urban intersections receives strong research interest in the field of automated intersection management. So far, mostly nonlearning algorithms like reservation or optimization-based ones were proposed to solve the underlying multi-agent planning problem. At the same time, automated driving functions for a single ego vehicle are increasingly implemented using machine learning methods. In this work, we build upon a previously presented graph-based scene representation and graph neural network to approach the problem using reinforcement learning. The scene representation is improved in key aspects by using edge features in addition to the existing node features for the vehicles. This leads to an increased representation quality that is leveraged by an updated network architecture. The paper provides an in-depth evaluation of the proposed method against baselines that are commonly used in automatic intersection management. Compared to a traditional signalized intersection and an enhanced first-in-first-out scheme, a significant reduction of traversal duration is observed at varying traffic densities. Finally, the generalization capability of the graphbased representation is evaluated by testing the policy on intersection layouts not seen during training. The model generalizes virtually without restrictions to smaller intersection layouts and within certain limits to larger ones.

Automatic Intersection Management in Mixed Traffic Using Reinforcement Learning and Graph Neural Networks

“…Our previous work [2] was the first to propose a GNN trained in RL for centralized cooperative behavior planning in fully automated traffic. The representation and learning model have been improved and was shown to generalize to intersection layouts not encountered during training [3].…”

“…We retain the core idea of encoding the traffic scene at an urban intersection as a directed graph with vertex features and edge features from [3]. Graph-based input representations proved to be well-suited for behavior planning in automated driving, where a varying number of dynamically interacting entities must be encoded efficiently [24].…”

“…We propose a GNN architecture that is composed of relational graph convolutional network (RGCN) layers [27] and graph attention (GAT) layers [28]. To process edge features in the RGCN layers, we use the extended update rule for message passing that was introduced in [3]. The different edge types correspond to independently learnable weight matrices.…”

“…In this work, we use an enhanced first in -first out (eFIFO) scheme as a baseline to the proposed RL planner. A more extensive set of baselines was employed in [3] for fully automated traffic, while we restrict the current analyses to the best performing baseline for conciseness. The algorithmic idea of the eFIFO is being extended to handle mixed traffic.…”

“…In the current work, we propose a mixed-traffic-capable cooperative behavior planning scheme using reinforcement learning (RL) and graph neural networks (GNNs). We build our approach on our previous work proposing a cooperative planning model for fully automated traffic [2], [3]. The contribution of the current work is fourfold:…”