Scientific Papers

Driving safety zone model oriented motion planning framework for autonomous truck platooning

The ever-increasing traffic demand poses a critical challenge to the existing transportation system, frequently resulting in traffic congestion and even fatalities (Zheng et al., 2018). Autonomous platooning is a highly anticipated solution for enhancing traffic capacity (Fernandes and Nunes, 2012) and reducing energy consumption (Turri et al., 2017). It ensures that vehicles in the platoon maintain the desired speed with a limited inter-vehicle distance. However, the limited inter-vehicle distance poses a significant threat to the safety of autonomous platoons especially for heavy-duty transportation (Turri et al., 2017).

Besides, for the implication of the autonomous platoon, there will be a heterogeneous driving environment consisting of both autonomous and human-driven vehicles. Given that a significant proportion of today’s traffic accidents are the results of human errors, a heterogeneous driving environment poses significant safety risks due to complex traffic interactions (Wang et al., 2020). Specifically in a heterogeneous driving environment, aggressive vehicle behaviors (unexpected cut in/out, emergency braking, illegal overtaking, etc.) from internal and external vehicles of the platoon will trigger dynamic perturbations and variations of inter-vehicle spaces, endangering autonomous platoon driving safety (Song et al., 2023).

For instance, Fig. 1 depicts a critical scenario for autonomous platooning in a heterogeneous driving environment, in which both internal and external vehicles of the platoon engage in aggressive vehicle behaviors. An autonomous platoon consisting of 4 commercial vehicles is operating on the right lane of the highway while two human-driving vehicles are on the left lane. Obstacle vehicle 1 abruptly crosses into lane 1 and decelerates as it tries to leave the highway at the next exit. Meanwhile, the leading vehicle of the autonomous platoon attempts to overtake obstacle vehicle 2 in lane 2 due to its slower velocity, which is against the law (overtaking in the right lane is prohibited) but is a common occurrence in real life. As a consequence of the series of unsafe driving behaviors, the leading vehicle inevitably collides with obstacle vehicle 1 or changes to lane 1 and collides with obstacle vehicle 2. Additionally, the rest of the platoon will experience critical dynamic perturbations and may also be involved in accidents. Therefore, in a complex heterogeneous driving environment, autonomous platoons must acquire capabilities to make safe and effective behavioral decisions and motion planning.

For quantifying and mitigating crash risks while autonomous vehicles interact with other traffic participants within complex operational design domains (ODD), the safety guarantee model of autonomous vehicles is in demand. Continuous research efforts have been devoted to the development of safety guarantee models for autonomous vehicles (Wang et al., 2019, Yang et al., 2021, Wang et al., 2021, Liu et al., 2022). The conventional concept of autonomous vehicle safety is the prevention of accidents and losses. Hence, autonomous vehicle safety measures are first proposed based on the occurrence of accidents. Numerous researchers from both industry and academia have cited the mileage-testing method to address the safety liability of autonomous driving systems (Lv et al., 2018). Google, for example, annually analyses the failure incidents of autonomous vehicles and investigates whether they would have resulted in traffic accidents, and reports its findings to the Department of Motor Vehicles (DMV) (Google Auto). However, mileage testing alone may not cover edge cases including unpredictable events, complex traffic situations or challenging road conditions (Yang et al., 2023). Moreover, sufficient data cannot be provided to thoroughly assess the vehicle’s behavior in such situations and refine the autonomous driving systems to ensure safety (Ma et al., 2022). Besides, during mileage testing, there is always the risk of accidents or incidents on public roads, which can pose safety hazards to both the autonomous vehicle and other road users (Kalra and Paddock, 2016). Thus, the results from mileage testing cannot guarantee a reliable autonomous driving system.

As an alternative, MobilEye develops the Responsibility Sensitive Safety (RSS) model to standardize the quantitative and qualitative driving safety of autonomous vehicles (Shalev-Shwartz et al., 2018). Moreover, driving safety is interpreted as mathematical model that measures and defines whether an autonomous vehicle could operate safely despite the unsafe driving behaviors of other traffic participants (Xu et al., 2021). In addition, Gassmann (Gassmann et al., 2019) proposes an open-source C++ library for the RSS model. It can be integrated with existing autonomous driving software to validate and verify the safety of autonomous vehicles. However, the RSS model relies on a simplified representation of scenarios, assuming idealized road conditions and simplified vehicle dynamics. In terms of autonomous platooning, multiple vehicles are operating in close proximity, which introduces complexities in terms of coordination, communication, and dynamic interactions. Moreover, autonomous platoons can exhibit emergent behaviors such as merging, lane changing or cooperative maneuvers that are not explicitly defined in the RSS model. Hence, the RSS model which is designed for safety measurements of a single autonomous vehicle, may not be suitable for direct implementation on autonomous platoons.

The platooning safety of intelligent connected vehicles has also been investigated for several decades (Axelsson, 2017). According to the automotive-specific ISO 21448 (International Organization for Standardization, 2022), conventional methods such as Hazards and Risk Analysis (HARA), Fault Tree Analysis (FTA), and System Theoretical Process Analysis (STPA) were used to identify potential hazards which could lead to system-level failure. For example, the EU project ENSEMBLE identifies a variety of critical scenarios involving infrastructures, human factors, hardware and software malfunctions, etc. (Mascalchi & Willemsen, 2020). To address the specific component-level safety of the platoon system, the majority of previous research has focused on control theory and game theory (Zhou et al., 2020, Lan and Zhao, 2020, Alam et al., 2014, Hong et al., 2020, Alam et al., 2015, Wang et al., 2022, Corno et al., 2023). For instance, to prevent internal collisions, string stability is an essential concept that indicates the magnitude of disturbances that should be suppressed from the leading vehicle to the following vehicles. Zhou et al. proposed an H-infinity controller to ensure string stability for a mixed platoon of vehicles (Zhou et al., 2020). Moreover, a min–max model predictive controller that sets the l2-norm in the cost function was proposed in (Lan and Zhao, 2020) to ensure the string stability of the platoon in the presence of V2X equipment’s communication delay. To define the proper safe inter-vehicle distance, a safe reachability set computation approach is proposed in (Alam et al., 2014). The platoon system is formulated as a pursuit-evasion game, and collision-causing unsafe reachability sets are calculated. Several mathematical safety evaluation indices, such as time-exposed time-to-collision, time-integrated time-to-collision, and time exposed rear-end crash risk index, are defined in (Rahman and Abdel-Aty, 2018) using a surrogate safety assessment technique. However, the vast majority of the existing literature interprets platooning safety exclusively in terms of longitudinal dynamics. When the platoon encounters complex situations, such as vehicles cutting in and out, the narrow consideration of the longitudinal safety of an autonomous platoon is not sufficient. Meanwhile, the validation scenarios which only consider the interactions within the platoon, are quite idealistic (Ge et al., 2022, Shen et al., 2022, Thormann et al., 2022). In this sense, a comprehensive platooning safety guaranteed model which takes longitudinal and lateral movements into account remains a blank field to be investigated.

Relevant prior work is examined in the following section. The motion control framework (MCF) designed in (Wang et al., 2023) utilized the H-infinity algorithm, which generated the longitudinal referenced velocity, and the Model Predictive Controller, which executed control commands based on the referenced signals, Artificial Potential Field, and vehicle dynamic constraints. However, the platoon’s overtaking maneuver is aggressive and occasionally increases operational risks, particularly when it interacts with multiple traffic participants simultaneously. The reason for this is that the platoon is passively avoiding obstacles to minimize the potential field value while ignoring the intentions of other traffic participants.

The above discussion demonstrates that the conventional mileage testing and the RSS model are not sufficient to assess the safety evaluation of autonomous platoons. Besides, platooning safety cannot be comprehensively guaranteed by narrow considerations of longitudinal movements. Consequently, this paper presents a novel driving safety zone-oriented motion planning framework (DSZMF) to interact rationally with surrounding vehicles and resolve the safety guaranteed control issues for the autonomous platoon.

Source link