Monte-Carlo Tree Search with Neural Network Guidance for Lane-Free Autonomous Driving

TL;DR

This work addresses planning for autonomous driving in lane-free traffic by integrating Monte-Carlo Tree Search (MCTS) with a pre-trained neural network (NN) to guide the selection process under computational constraints. The authors formulate a lane-free driving MDP and incorporate NN priors into the MCTS via a PUCT-style score, trained through self-play and calibrated with reliability diagrams to ensure well-behaved predictions. They introduce a nudging mechanism by extending state information to include rear vehicles, enabling proactive safety behavior. Experiments across three traffic densities show that NN-guided MCTS accelerates convergence and improves safety and speed trade-offs compared to plain MCTS and a standalone NN, highlighting the practical benefits of neural guidance under realistic time budgets.

Abstract

Lane-free traffic environments allow vehicles to better harness the lateral capacity of the road without being restricted to lane-keeping, thereby increasing the traffic flow rates. As such, we have a distinct and more challenging setting for autonomous driving. In this work, we consider a Monte-Carlo Tree Search (MCTS) planning approach for single-agent autonomous driving in lane-free traffic, where the associated Markov Decision Process we formulate is influenced from existing approaches tied to reinforcement learning frameworks. In addition, MCTS is equipped with a pre-trained neural network (NN) that guides the selection phase. This procedure incorporates the predictive capabilities of NNs for a more informed tree search process under computational constraints. In our experimental evaluation, we consider metrics that address both safety (through collision rates) and efficacy (through measured speed). Then, we examine: (a) the influence of isotropic state information for vehicles in a lane-free environment, resulting in nudging behaviour--vehicles' policy reacts due to the presence of faster tailing ones, (b) the acceleration of performance for the NN-guided variant of MCTS, and (c) the trade-off between computational resources and solution quality.

Figures (21)

• Figure 1: Illustration of NN-Guided selection, where the $child_i$ with the highest PUCT value is selected.
• Figure 2: Snapshot of the simulation environment.
• Figure 3: Action Space consists of $15$ discrete actions in the $2$-dimensional space $\{a_x, a_y\}$.
• Figure 4: Reliability Diagram of trained NN's calibration compared to a perfect calibration plotted line.
• Figure 5: Graph of Collisions$\pm$SD for NN-MCTS, MCTS (nudging), MCTS and NN across different Iterations for 5400 ${veh/h}$.