VLM-RL: A Unified Vision Language Models and Reinforcement Learning Framework for Safe Autonomous Driving

强化学习 计算机科学 钢筋 人工智能 人机交互 心理学 社会心理学
作者
Zilin Huang,Zihao Sheng,Yansong Qu,Junwei You,Sikai Chen
出处
期刊:Cornell University - arXiv [Cornell University]
标识
DOI:10.48550/arxiv.2412.15544
摘要

In recent years, reinforcement learning (RL)-based methods for learning driving policies have gained increasing attention in the autonomous driving community and have achieved remarkable progress in various driving scenarios. However, traditional RL approaches rely on manually engineered rewards, which require extensive human effort and often lack generalizability. To address these limitations, we propose \textbf{VLM-RL}, a unified framework that integrates pre-trained Vision-Language Models (VLMs) with RL to generate reward signals using image observation and natural language goals. The core of VLM-RL is the contrasting language goal (CLG)-as-reward paradigm, which uses positive and negative language goals to generate semantic rewards. We further introduce a hierarchical reward synthesis approach that combines CLG-based semantic rewards with vehicle state information, improving reward stability and offering a more comprehensive reward signal. Additionally, a batch-processing technique is employed to optimize computational efficiency during training. Extensive experiments in the CARLA simulator demonstrate that VLM-RL outperforms state-of-the-art baselines, achieving a 10.5\% reduction in collision rate, a 104.6\% increase in route completion rate, and robust generalization to unseen driving scenarios. Furthermore, VLM-RL can seamlessly integrate almost any standard RL algorithms, potentially revolutionizing the existing RL paradigm that relies on manual reward engineering and enabling continuous performance improvements. The demo video and code can be accessed at: https://zilin-huang.github.io/VLM-RL-website.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
海宁完成签到,获得积分10
刚刚
瘦瘦的南蕾完成签到,获得积分20
1秒前
虎啸山河完成签到,获得积分10
1秒前
1秒前
张艳完成签到,获得积分20
1秒前
liujing_242022完成签到,获得积分10
1秒前
2秒前
xiaobai完成签到,获得积分10
2秒前
斯文败类的应助被chen采纳,获得30
2秒前
慕青的应助被君尧采纳,获得30
2秒前
科研狗的应助被七月流火采纳,获得50
2秒前
OeO发布了新的文献求助10
3秒前
NexusExplorer的应助被33988采纳,获得10
3秒前
斯文败类的应助被33988采纳,获得10
3秒前
桐桐的应助被33988采纳,获得10
3秒前
FashionBoy的应助被33988采纳,获得10
3秒前
科研通AI6.4的应助被eatme采纳,获得30
3秒前
海宁发布了新的文献求助10
4秒前
丘比特的应助被畅快山兰采纳,获得10
4秒前
6秒前
单纯雪柳完成签到,获得积分10
6秒前
6秒前
lalala完成签到,获得积分10
6秒前
6秒前
7秒前
maxiaole的应助被段棋晔采纳,获得10
8秒前
weimin完成签到,获得积分10
8秒前
超困困困狗完成签到,获得积分10
9秒前
Jasper的应助被33988采纳,获得10
9秒前
小二郎的应助被33988采纳,获得10
9秒前
bkagyin的应助被33988采纳,获得10
9秒前
科研助理795的应助被33988采纳,获得10
9秒前
传奇3的应助被33988采纳,获得10
9秒前
9秒前
10秒前
万能图书馆的应助被33988采纳,获得10
10秒前
可爱的函函的应助被33988采纳,获得10
10秒前
科目三的应助被33988采纳,获得10
10秒前
sdl发布了新的文献求助10
10秒前
完美世界的应助被33988采纳,获得10
10秒前
高分求助中
(应助此贴封号)通过应助OA文献获取积分 10000
Rosenblum, Global Change Biology 800
Organizational Behavior 510
Management and the Arts 510
Convergent and bidirectional strategies towards the total synthesis of hemibrevetoxin B 300
Geschichtliche Grundbegriffe (GGB), Band 5: Pro–Soz 300
Die Religion in Geschichte und Gegenwart (RGG), 4. Auflage, Band 7: R–S 300
热门求助领域 (近24小时)
化学 材料科学 医学 生物 计算机科学 工程类 纳米技术 内科学 物理 有机化学 化学工程 生物化学 复合材料 光电子学 细胞生物学 心理学 量子力学 催化作用 物理化学 电极
热门帖子
关注 科研通微信公众号,转发送积分 7799435
求助须知:如何正确求助?哪些是违规求助? 9334501
关于积分的说明 20467379
捐赠科研通 7390524
什么是DOI,文献DOI怎么找? 3326026
关于科研通互助平台的介绍 2473113
邀请新用户注册赠送积分活动 2343529