基于语言推理和认知记忆的自动驾驶决策模型

doi:10.13229/j.cnki.jdxbgxb.20240606

摘要/Abstract

摘要：

针对传统自动驾驶系统安全性能不足、学习效率低等问题，提出了一种可持续学习和理解语言信息的自动驾驶安全决策模型。该模型参考人类驾驶的推理决策和经验积累过程，以大型语言模型（LLM）作为决策智能体，将思维链推理、两阶段注意力机制和认知记忆存储与检索整合到驾驶过程的上下文安全学习中；同时，采用运动学模块将LLM决策转化为可操作的驾驶命令，实现安全驾驶经验的持续学习。实验结果表明，本文决策模型相较于基于规则、强化学习和知识的方法，在安全、效率方面有显著提升，并具备持续学习和根据人类指令调整驾驶行为的能力，可为类人自动驾驶提供参考。

关键词: 车辆工程, 自动驾驶, 持续学习, 大语言模型, 思维链推理, 两阶段注意力机制

Abstract:

To address the issues of insufficent safety performance and low learning inefficient in traditional autonomous driving systems， an autonomous driving safety decision-making model capable of continuous learning and understanding linguistic information was proposed. Referring to the reasoning decision-making and experience accumulation processes in human driving， this model leverages a large language model （LLM） as the decision-making agent， integrating chain-of-thought reasoning， a two-stage attention mechanism， and cognitive memory storage and retrieval into the contextual safety learning of the driving process. Meanwhile， a kinematic module is employed to convert LLM decisions into executable driving commands， enabling the continuous learning of safe driving experiences. Experimental results demonstrate that the proposed decision-making model significantly improves safety and efficiency compared to rule， reinforcement learning， and knowledge-based approaches， and possesses the capability of continuous learning and adapting driving behaviors based on human instructions， providing a reference for human-like autonomous driving.

Key words: vehicle engineering, autonomous driving, continuous learning, large language model, chain-of-thought reasoning, two-stage attention mechanism

中图分类号:

U495

王祥,谭国真,彭衍飞,任浩,李健平. 基于语言推理和认知记忆的自动驾驶决策模型[J]. 吉林大学学报(工学版), 2025, 55(12): 3918-3927.

Xiang WANG,Guo-zhen TAN,Yan-fei PENG,Hao REN,Jian-ping LI. Autonomous driving decision⁃making model based on language reasoning and cognitive memory[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(12): 3918-3927.

图/表 14

图1

图2

图3

图4

图5

表1

表2

表3

图6

图7

图8

表4

表5

不同指令下的驾驶行为"

指导命令	$a ˉ$ /（m·s^-2）	$s ˉ$ /rad	$v ˉ$ /（m·s^-1）	$g ˉ$ /m	$l ˉ$
DMA	3.50	0.03	34.7	8.6	7
DMC	0.21	0.01	21.2	41.8	1
NEC	1.87	0.02	28.3	25.6	2

表5

图9

参考文献 15

[1]	马依宁, 姜为, 吴靖宇, 等. 基于不同风格行驶模型的自动驾驶仿真测试自演绎场景研究[J]. 中国公路报, 2023, 36(2): 216-228.
	Ma Yi-ning, Jiang Wei, Wu Jing-yu, et al. Self- evolution scenarios for simulation tests of autonomous vehicles based on different models of driving styles[J]. China Journal of Highway and Transport, 2023, 36(2): 216-228.
[2]	李伟东, 马草原, 史浩, 等. 基于分层强化学习的自动驾驶决策控制算法[J]. 吉林大学学报: 工学版, 2025, 55(5): 1798-1805.
	Li Wei-dong, Ma Cao-yuan, Shi Hao, et al. An automatic driving decision control algorithm based on
	hierarchical reinforcement learning[J]. Journal of Jilin University (Engineering and Technology Edition), 2025, 55(5): 1798-1805.
[3]	朱波, 张纪伟, 谈东奎, 等. 基于多源传感器与导航地图的端到端自动驾驶方法[J]. 汽车安全与节能学报, 2022, 13(4): 738-749.
	Zhu Bo, Zhang Ji-wei, Tan Dong-kui, et al. End-to-end autonomous driving method based on multi-source sensor and navigation map[J]. Journal of Automotive Safety and Energy, 2022, 13(4): 738-749.
[4]	Zhang Q X, Zhao Y H, Wang Y J, et al. Towards cross-task universal perturbation against black-box object detectors in autonomous driving[J]. Computer Networks, 2020, 180: No.107388.
[5]	Wang S Y, Zhu Y X, Li Z H, et al. ChatGPT as your vehicle co-pilot: An initial attempt[J]. IEEE Transactions on Intelligent Vehicles, 2023, 8(12): 4706-4721.
[6]	Cui Y D, Huang S C, Zhong J M, et al. DriveLLM: charting the path toward full autonomous driving with large language models[J]. IEEE Transactions on Intelligent Vehicles, 2023, 9(1): 1450-1464.
[7]	Kojima T, Gu S S, Reid M, et al. Large language models are zero-shot reasoners[J]. Advances in Neural Information Processing Systems, 2022, 35: 22199- 22213.
[8]	王祥, 谭国真. 基于知识与大语言模型的高速环境自动驾驶决策研究[J]. 系统仿真学报, 2025(5): 1246-1255.
	Wang Xiang, Tan Guo-zhen. Research on decision-making of autonomous driving in highway environment based on knowledge and large language model[J]. Journal of System Simulation, 2025(5): 1246-1255.
[9]	Peng Y F, Tan G Z, Si H W, et al. DRL-GAT-SA: Deep reinforcement learning for autonomous driving planning based on graph attention networks and simplex architecture[J]. Journal of Systems Architecture, 2022, 126: No.102505.
[10]	胡宏宇, 张慧珺, 姚荣涵, 等. L3级自动驾驶接管过程驾驶员情景意识研究[J]. 吉林大学学报: 工学版, 2024, 54(2): 410-418.
	Hu Hong-yu, Zhang Hui-jun, Yao Rong-han, et al. Driver's situational awareness in takeover process of L3 automated vehicles[J]. Journal of Jilin University (Engineering and Technology Edition), 2024, 54(2): 410-418.
[11]	Nie X T, Liang Y P, Ohkura K. Autonomous highway driving using reinforcement learning with safety check system based on time-to-collision[J]. Artificial Life and Robotics, 2023, 28(1): 158-165.
[12]	Chang M K, Lee S H, Chung C C. Comparative evaluation of dynamic and kinematic vehicle models[C]∥Conference on Decision and Control, Los Angeles, CA, USA, 2015: 648-653.
[13]	Treiber M, Hennecke A, Helbing D. Congested traffic states in empirical observations and microscopic simulations[J]. Physical Review E, 2000, 62(2): 1805.
[14]	Xin L, Kong Y T, Li S E, et al. Enable faster and smoother spatio-temporal trajectory planning for autonomous vehicles in constrained dynamic environment[J].Journal of Automobile Engineering, 2021, 235(4): 1101-1112.
[15]	Li G F, Li S L, Li S, et al. Deep reinforcement learning enabled decision-making for autonomous driving at intersections[J]. Automotive Innovation, 2020, 3: 374-385.

相关文章 15

[1]	兰巍,周政,王冠宇,王伟,张苗苗. 基于机器学习的汽车设计智能拟合方法[J]. 吉林大学学报(工学版), 2025, 55(9): 2858-2863.
[2]	孙天骏,杨惠喆,蔡荣贵,冯嘉仪,冉锐,刘斌. 面向纯电动汽车自适应巡航系统的人性化起停控制策略[J]. 吉林大学学报(工学版), 2025, 55(9): 2847-2857.
[3]	李寿涛,贾湘怡,朱军,郭洪艳,于丁力. 基于Level-K的智能驾驶汽车无信控交叉路口决策方法[J]. 吉林大学学报(工学版), 2025, 55(9): 3069-3078.
[4]	朱冰,孟鹏翔,刘斌,韩嘉懿,赵健,陈志成,宋东鉴,陶晓文. 基于交通环境信息的虚拟车道线拟合方法[J]. 吉林大学学报(工学版), 2025, 55(9): 2935-2945.
[5]	赵俊武,曲婷,胡云峰. 基于自适应采样的智能车辆轨迹规划方法[J]. 吉林大学学报(工学版), 2025, 55(8): 2802-2816.
[6]	于贵申,陈鑫,唐悦,赵春晖,牛艾佳,柴辉,那景新. 激光表面处理对铝-铝粘接接头剪切强度的影响[J]. 吉林大学学报(工学版), 2025, 55(8): 2555-2569.
[7]	高金武,孙少龙,王舜尧,高炳钊. 基于电机转矩补偿的增程器转速波动抑制策略[J]. 吉林大学学报(工学版), 2025, 55(8): 2475-2486.
[8]	贾美霞,胡建军,肖凤. 基于多软件联合的车用电机变工况多物理场仿真方法[J]. 吉林大学学报(工学版), 2025, 55(6): 1862-1872.
[9]	宋学伟,于泽平,肖阳,王德平,袁泉,李欣卓,郑迦文. 锂离子电池老化后性能变化研究进展[J]. 吉林大学学报(工学版), 2025, 55(6): 1817-1833.
[10]	肖纯,易子淳,周炳寅,张少睿. 基于改进鸽群优化算法的燃料电池汽车模糊能量管理策略[J]. 吉林大学学报(工学版), 2025, 55(6): 1873-1882.
[11]	王健,贾晨威. 面向智能网联车辆的轨迹预测模型[J]. 吉林大学学报(工学版), 2025, 55(6): 1963-1972.
[12]	李伟东,马草原,史浩,曹衡. 基于分层强化学习的自动驾驶决策控制算法[J]. 吉林大学学报(工学版), 2025, 55(5): 1798-1805.
[13]	卢荡,索艳茹,孙宇航,吴海东. 基于无量纲格式的轮胎侧倾侧偏力学特性预测[J]. 吉林大学学报(工学版), 2025, 55(5): 1516-1524.
[14]	高镇海,郑程元,赵睿. 真实与虚拟场景下自动驾驶车辆的主动安全性验证与确认综述[J]. 吉林大学学报(工学版), 2025, 55(4): 1142-1162.
[15]	张涛,林黄达,余中军. 混合动力车辆换挡的实时滚动优化控制方法[J]. 吉林大学学报(工学版), 2025, 55(4): 1215-1224.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

数据流	参数	类型	说明
输入	model	string	llama2-7b-chat-v2
	messages	list	输入的内容
	result_format	string	用户返回的内容类型
输出	request_id	string	llama2-7b-chat-v2
	output	list	调用结果信息
	usage.input_ tokens	int	用户输入文本转换为Token后的长度
	usage.output_ tokens	int	模型生成回复转换为Token后的长度

结构	取值
输入	［·，6］
层数	编码器［64，64］；2个注意力头；d_k =32；解码器［64，64］
参数量	3.4×10⁴

场景	方法	C	ζ	t	P_D	P_d	C_n
高速环境	LRCMM	0	12.8	26.4	1.35	1.31	13.89
	RBM	4	26.8	19.7	3.25	3.13	27.83
	MFRLM	8	35.2	15.7	3.84	3.45	46.95
	KDM	3	14.6	29.5	2.47	2.31	20.84
十字路口	LRCMM	0	32.8	47.2	1.84	0.91	24.22
	RBM	2	75.1	32.1	4.21	2.64	48.11
	MFRLM	9	68.4	30.5	5.28	1.06	50.50
	KDM	7	42.5	58.5	3.98	3.61	40.31

场景	TAM	CM	C	ζ	t	P_D	P_d	C_n
高速环境	×	×	6	69.5	34.3	4.35	3.94	48.00
	√	×	4	64.7	36.5	3.84	3.12	43.82
	×	√	3	38.6	28.4	2.37	1.26	26.90
	√	√	0	14.7	27.1	1.44	1.29	14.97
十字路口	×	×	2	33.7	48.9	4.21	3.95	38.03
	√	×	3	32.5	47.8	4.18	3.84	37.35
	×	√	1	32.1	47.3	3.89	3.52	35.57
	√	√	0	31.7	46.2	3.74	3.43	34.55