Path tracking is an essential aspect of the navigational process of self-driving cars. Especially, accurate path tracking is important for not only normal urban roads but also narrow and complex roads such as parking lot and alleyway. The pure pursuit method is one of the geometric path-tracking methods. Using this method, the look-ahead point can be selected far away and the control input is computed in real-time, which is advantageous when the given path is not smooth or when the path is specified using waypoints. Moreover, this method is more robust to localization errors than the model-based path-tracking method. However, the original pure pursuit method and its variants have limited tracking performance. Therefore, this paper proposes a new method that heuristically selects a look-ahead point by considering the relationship between a vehicle and a path. Using this new look-ahead point, the vehicle can stably converge to the desired path and track the path without encountering the cutting-corner problem. The proposed method was tested using simulation and our self-driving car platform. Our results show that the vehicle tracks the desired path more accurately using our proposed algorithm than using the previous pure pursuit methods.

路径跟踪是自动驾驶汽车导航过程的重要方面。尤其是，精确的路径跟踪不仅对于正常的城市道路，而且对于狭窄而复杂的道路（例如停车场和小巷）都非常重要。纯追踪方法是几何路径跟踪方法之一。使用此方法，可以选择远处的超前点并实时计算控制输入，这在给定路径不平滑或使用航路点指定路径时非常有利。此外，与基于模型的路径跟踪方法相比，该方法对定位错误的鲁棒性更高。但是，原始的纯追踪方法及其变体具有有限的跟踪性能。因此，本文提出了一种新方法，该方法通过考虑车辆与路径之间的关系来启发式选择前瞻点。使用这个新的预视点，车辆可以稳定地收敛到所需的路径并跟踪该路径，而不会遇到切角问题。使用仿真和我们的自动驾驶汽车平台对提出的方法进行了测试。我们的结果表明，使用我们提出的算法比使用以前的纯追踪方法，车辆可以更准确地跟踪所需的路径。