驾驶风格研究：基于逆强化学习

34 浏览量更新于2024-08-29 收藏 512KB PDF 举报

"研究表明，基于逆强化学习的驾驶者驾驶风格分析" 在当今的汽车工业中，先进驾驶辅助系统（ADAS）已经广泛应用，旨在提升驾驶安全性、舒适度，并减轻驾驶员的驾驶负担。然而，尽管这些系统功能强大，但它们通常并未考虑到不同驾驶员的个性化驾驶风格。这一点对于提供舒适的驾驶体验和提高市场接受度至关重要。理解并识别驾驶者的驾驶风格是一项挑战，因为驾驶人群数量庞大且行为差异显著。过去的科研工作主要采用物理方法来建模驾驶员的驾驶行为，但这种方法往往过于简化，无法全面捕捉到驾驶者的行为细节和多样性。针对这一问题，"基于逆强化学习的驾驶者驾驶风格研究"这篇论文提出了一种新的方法。逆强化学习是一种机器学习技术，它能从观察到的行为推断出决策者的目标或策略，这在理解复杂的、非结构化的人类行为如驾驶风格时尤其有用。在该研究中，作者Yuande Jiang, Weiwen Deng (吉林大学), Jinsong Wang (通用汽车公司), 和 Bing Zhu (吉林大学)运用逆强化学习技术，试图从大量驾驶员的行为数据中学习和理解他们的驾驶习惯和偏好。通过这种方法，系统可以模拟不同的驾驶风格，并可能实现对ADAS的个性化定制，使其能够适应每个驾驶员的独特风格。这项工作的创新之处在于，它不仅关注驾驶行为的物理特性，如速度、加速度等，还深入探究了影响驾驶风格的心理和社会因素。通过逆强化学习，可以揭示驾驶员在特定情境下做出决策的潜在动机，进一步帮助设计更智能、更人性化的ADAS系统。论文的结论指出，逆强化学习为理解和模拟驾驶员驾驶风格提供了新的视角，有望推动ADAS技术的发展，使未来的车辆能够更好地适应并响应驾驶员的需求，从而提高驾驶的舒适性和安全性。同时，这种个性化的驾驶辅助系统也可能有助于减少因驾驶风格差异引起的交通事故，提升整体的道路安全水平。这篇2018年的SAE Technical Paper 2018-01-0612通过逆强化学习技术，为驾驶行为研究开辟了新的道路，为ADAS的未来发展提供了理论支持和实践指导。

2018-01-0612 Published 03 Apr 2018

Studies on Drivers’ Driving Styles Based on Inverse

Reinforcement Learning

Yuande Jiang and Weiwen Deng Jilin University

Jinsong Wang General Motors LLC

Bing Zhu Jilin University

Citation: Jiang, Y., Deng, W., Wang, J., and Zhu, B., “Studies on Drivers’ Driving Styles Based on Inverse Reinforcement Learning,”

SAETechnical Paper 2018-01-0612, 2018, doi:10.4271/2018-01-0612.

Abstract

lthough advanced driver assistance systems (ADAS)

have been widely introduced in automotive industry

to enhance driving safety and comfort, and to reduce

drivers’ driving burden, they do not in general reect dierent

drivers’ driving styles or customized with individual person-

alities. This can be important to comfort and enjoyable

driving experience, and to improved market acceptance.

However, it is challenging to understand and further identify

drivers’ driving styles due to large number and great varia-

tions of driving population. Previous research has mainly

adopted physical approaches in modeling drivers’ driving

behavior, which however are oen very much limited, if not

impossible, in capturing human drivers’ driving character-

istics. is paper proposes a reinforcement learning based

approach, in which the driving styles are formulated through

drivers’ learning processes from interaction with surrounding

environment. Based on the reinforcement learning theory,

driving action can be treated as maximizing a reward

function. Instead of calibrating the unknown reward function

to satisfy driver’s desired response, we try to recover it from

the human driving data, utilizing maximum likelihood

inverse reinforcement learning (MLIRL). An IRL-based longi-

tudinal driving assistance system is also proposed in this

paper. Firstly, large amount of real world driving data is

collected from a test vehicle, and the data is split into two sets

for training and for testing purposes respectively. en, the

longitudinal acceleration is modeled as a Boltzmann distribu-

tion in human driving activity. The reward function is

denoted as a linear combination of some kernelized basis

functions. e driving style parameter vector is estimated

using MLIRL based on the training set. Finally, a learning-

based longitudinal driving assistance algorithm is developed

and evaluated on the testing set. e results demonstrate that

the proposed method can satisfactorily reect human drivers’

driving behavior.

Introduction

n past decades, great progress has been acheived in the

development of various technology areas, such as sensing,

computer technology, embedded system and digital control

technology, which has further drived the advance develop-

ment for intelligent driving systems. Advanced driver assis-

tance systems (ADAS) are examples which have gained wide

applications to improve driving safety and comfort. Some

highly automated driving technologies are further on the way

to market. Autonomous driving technologies also become

frontier areas in academia and industry research. Along with

the advanced development in autonomous driving, the study

of the interaction between human (or human driver) and

machine (or intelligent systems) is becoming

increasingly prominent.

Currently, some research activities have been conducted

to take driver’s preferences and driving characteristics into

account to improve the performance of intelligent driving

systems. According to dierent purposes of use, these methods

can be classied into two categories: personalized assistance

system design, and estimation of the likely behavior of human

driven vehicle for autonomous vehicle. In ADAS, driver’s char-

acteristics are the most complicated factors which have great

impacts on the acceptance of these systems. In the design of

personalized driver assistance systems, driver’s characteristics

are considered mainly by using model-based approach and

learning-based approach. In model-based approaches, driver

behavior is modeled with a xed structure. For example, the

car-following process is treated as a linear regression function

of several typical features in [1, 2, 3, 4], and dierent driving

characteristics are represented by the model parameters. Some

nonlinear models are proposed in [5, 6, 7], and a probability

weighted autoregressive exogenous (PWARX) model, an

extension of normal linear mode, is introduced in [8, 9]. In

addition, some studies assume that individual driver charac-

teristics are the results of trade-o among some control objec-

tives [10, 11]. It can be found that the model-based approach

is implemented based on an underlying assumption that

human driving process can be modeled physically to some

extent. However, due to its inherent complexity and

Downloaded from SAE International by Jilin University, Saturday, January 04, 2020

下载后可阅读完整内容，剩余7页未读，立即下载

weixin_38554781

粉丝: 6
资源: 884

驾驶风格研究：基于逆强化学习

Inverse Reinforcement Learning.pptx

inverse reinforcement learning

Nonlinear Inverse Reinforcement Learning with Mutual Information and Gaussian Process

Inverse-Reinforcement-Learning:选定的逆强化学习算法的实现

A neural networks based model of inverse hysteresis

A simple data assimilation method for improving the MODIS LAI time-series data products based on the object analysis and gradient inverse weighted filter

lane-and-obstacle-detection-based-on-fast-inverse-perspective-ma.pdf

Prediction of electromagnetic field distributions inside biological bodies by using an inverse scattering procedure based on a statistical cooling algorithm

Inverse displacement analysis of the general six degree-of-freedom serial robot based on optimization method (2011年)

Cumulant-Based-Inverse-Filter-Criteria.rar_matlab例程_matlab_

最新资源