Abstract: With the continuous development of artificial intelligence and autonomous driving technology, autonomous driving technology based on electrification, networking, intelligence, and sharing has become the main development direction of the future automotive industry. However, due to the limitations of current autonomous driving technology and relevant policies and regulations, it can be foreseen that human-machine co-driving will exist as a transitional stage towards autonomous driving for a long time. Therefore, this paper proposes a machine learning method based on convolutional neural networks to address the issue of driver operation intention recognition in human-machine co-driving. Starting from the steering wheel that the driver has the longest contact with, the problem of operation intention recognition is transformed into a problem of image classification and recognition through feature training of tactile pressure images, accurate identification of different driving intentions of drivers (grip driving intentions, grip driving intentions, dangerous driving intentions) has been achieved. The simulation results show that the accuracy of intention recognition exceeds 97%, which has important theoretical and practical value for the future research of vehicle assisted driving systems in human-machine co-driving control.

Key words: Autonomous driving; Human-machine co-driving; Intention recognition; Tactile information; Machine learning

摘要： 随着人工智能和自动驾驶技术的不断发展，以电动化、网联化、智能化和共享化为基 础的自动驾驶技术已成为未来汽车工业的主要发展方向。然而，受当前自动驾驶技术和相关 政策法规制定的限制，可以预见，人机共驾将作为面向自动驾驶的过度阶段将长时间存在。 因此，针对人机共驾中驾驶员操作意图识别问题，文章提出了一种基于卷积神经网络的机器 学习方法，从驾驶员最长接触的转向盘入手，通过对触觉压力图像的特征训练，将操作意图 的识别问题转化为对图像的分类识别问题，实现了对不同驾驶意图（紧抓握行驶意图、松抓 握行驶意图、危险驾驶意图）的准确识别。仿真结果表明，意图识别准确率超过 97%，对将 来人机共驾控制中车辆辅助驾驶系统的研究具有重要的理论和应用价值。

关键词: 自动驾驶；人机共驾；意图识别；触觉信息；机器学习