A Survey of Reinforcement Learning-Based Motion Planning for Autonomous Driving: Lessons Learned from a Driving Task Perspective

TL;DR

This survey tackles RL-based motion planning for autonomous driving from a driving-task perspective, linking RL design choices to the specific characteristics and requirements of tasks such as car following, lane changing, ramp merging, intersections, parking, urban navigation, racing, and off-road driving. It synthesizes fundamentals of RL, MoP formulation, and task-driven design patterns, highlighting how observation inputs, action outputs, and reward functions must be tailored to each scenario. The paper also reviews frontier efforts addressing safety (CMDP, safe RL), sample efficiency (LfD, curriculum, transfer learning), and generalization (domain randomization, meta-learning, continual learning, LLMs), offering concrete guidance and identifying open challenges. By connecting methodological choices to driving tasks, the survey provides a practical roadmap for developing RL-based MoP solutions that balance safety, efficiency, and generalization in real-world AD systems.

Abstract

Reinforcement learning (RL), with its ability to explore and optimize policies in complex, dynamic decision-making tasks, has emerged as a promising approach to addressing motion planning (MoP) challenges in autonomous driving (AD). Despite rapid advancements in RL and AD, a systematic description and interpretation of the RL design process tailored to diverse driving tasks remains underdeveloped. This survey provides a comprehensive review of RL-based MoP for AD, focusing on lessons from task-specific perspectives. We first outline the fundamentals of RL methodologies, and then survey their applications in MoP, analyzing scenario-specific features and task requirements to shed light on their influence on RL design choices. Building on this analysis, we summarize key design experiences, extract insights from various driving task applications, and provide guidance for future implementations. Additionally, we examine the frontier challenges in RL-based MoP, review recent efforts to addresse these challenges, and propose strategies for overcoming unresolved issues.

Figures (5)

• Figure 1: Search result of Web of Science until 2024: (a) topic search for RL and AD. (b) topic search for surveys for RL, AD, and RL-based AD.
• Figure 2: The schematic of the survey structure of RL-based MoP for AD.
• Figure 3: RL methods with different categorization.
• Figure 4: RL algorithm applied to MoP for AD.
• Figure 5: Illustration of RL-based MoP for different driving tasks.