pdf

Multimodal perception parameterized decision making for autonomous driving

  • 2024-09-04
  • 2.34MB
  • Points it Requires : 1

针对自主驾驶的多模态感知参数化决策

Abstract—Autonomous driving is an emerging technology that

has advanced rapidly over the last decade. Modern transportation

is expected to benefit greatly from a wise decision-making

framework of autonomous vehicles, including the improvement of

mobility and the minimization of risks and travel time. However,

existing methods either ignore the complexity of environments

only fitting straight roads, or ignore the impact on surrounding

vehicles during optimization phases, leading to weak environmental adaptability and incomplete optimization objectives. To

address these limitations, we propose a pArameterized decisionmaking framework with mUlti-modal percepTiOn based on deep

reinforcement learning, called AUTO. We conduct a comprehensive perception to capture the state features of various traffic

participants around the autonomous vehicle, based on which we

design a graph-based model to learn a state representation of

the multi-modal semantic features. To distinguish between lanefollowing and lane-changing, we decompose an action of the

autonomous vehicle into a parameterized action structure that

first decides whether to change lanes and then computes an exact

action to execute. A hybrid reward function takes into account

aspects of safety, traffic efficiency, passenger comfort, and impact

to guide the framework to generate optimal actions. In addition,

we design a regularization term and a multi-worker paradigm to

enhance the training. Extensive experiments offer evidence that

AUTO can advance state-of-the-art in terms of both macroscopic

and microscopic effectiveness.

Index Terms—Decision-making, Autonomous Vehicle, Reinforcement Learning

unfold

You Might Like

Uploader
MartinFowler
 

Recommended ContentMore

Popular Components

Just Take a LookMore

EEWorld
subscription
account

EEWorld
service
account

Automotive
development
circle

About Us Customer Service Contact Information Datasheet Sitemap LatestNews


Room 1530, 15th Floor, Building B, No.18 Zhongguancun Street, Haidian District, Beijing, Postal Code: 100190 China Telephone: 008610 8235 0740

Copyright © 2005-2024 EEWORLD.com.cn, Inc. All rights reserved 京ICP证060456号 京ICP备10001474号-1 电信业务审批[2006]字第258号函 京公网安备 11010802033920号
×