SwapTransformer: highway overtaking tactical planner model via imitation learning on OSHA dataset

This paper investigates the high-level decision-making problem in highway scenarios regarding lane changing and over-taking other slower vehicles. In particular, this paper aims to improve the Travel Assist feature for automatic overtaking and lane changes on highways. About 9 million samples including lane images and other dynamic objects are collected in simulation. This data; Overtaking on Simulated HighwAys (OSHA) dataset is released to tackle this challenge. To solve this problem, an architecture called SwapTransformer is designed and implemented as an imitation learning approach on the OSHA dataset. Moreover, auxiliary tasks such as future points and car distance network predictions are proposed to aid the model in better understanding the surrounding environment. The performance of the proposed solution is compared with a multi-layer perceptron (MLP) and multi-head self-attention networks as baselines in a simulation environment. We also demonstrate the performance of the model with and without auxiliary tasks. All models are evaluated based on different metrics such as time to finish each lap, number of overtakes, and speed difference with speed limit. The evaluation shows that the SwapTransformer model outperforms other models in different traffic densities in the inference phase. In the past decade, the field of autonomous driving has received lots of attention. Self-driving cars or autonomous vehicles (AV) represent a novelty of artificial intelligence (AI), robotics, computer vision, and sensor technology Grigorescu et al. (2020); Xiao et al. (2020); Zablocki et al. (2022). Many works focused on end-to-end learning approaches from camera to direct actions such as steering wheel, acceleration, and break Bojarski et al. (2016); Kim & Park (2017); Yang et al. (2018); however, there are many challenges such as lack of interpretability, data efficiency, safety and robustness, generalization, and trade-off between layers that make the end-to-end training less suitable for self-driving cars reliability. On the other hand, modular approaches break down the problem into different tasks such as perception and sensor cognition, motion prediction, high-level and low-level path planner, and motion controller Grigorescu et al. (2020); Atakishiyev et al. (2021); Teng et al. (2023).