Design and field evaluation of cooperative adaptive cruise control with unconnected vehicle in the loop
Introduction
In order to improve traffic safety and mobility, advanced technologies have been introduced and implemented through Intelligent Transportation Systems (ITS). Connected Automated Vehicle (CAV), which enables a variety of cooperative automated driving applications (Shladover, 2018), is expected to play a key role in ITS. One promising CAV application is Cooperative Adaptive Cruise Control (CACC) (Naus et al., 2010), which is developed from Adaptive Cruise Control (ACC). In addition to measuring the position of the preceding vehicle by onboard sensor, the Vehicle-to-Vehicle (V2V) communications enables CACC vehicles to immediately obtain the traffic situation in the downstream, which helps improve the precision of longitudinal motion control (Milanes et al., 2014). As a result, the equipped vehicles can maintain short inter-vehicle headways (e.g., 0.6 s) with guaranteed safety and “string stability.”
String stability (Naus et al., 2010) refers to the vehicle’s capability of attenuating traffic disturbance from downstream, and keeping the vehicle string undisrupted by sudden acceleration or deceleration of any vehicles ahead of the platoon. In string-unstable situations, the speed/spacing disturbance from the preceding vehicle is amplified by the following vehicles, causing shockwaves along the traffic. It has been pointed out that the string-unstable dynamics of human-driven vehicles is an origin of the traffic jams that occur without apparent external reason (Sugiyama et al., 2008). Such disruptive behaviors were also observed on the existing commercial ACC systems (Gunter et al., 2020). Therefore, the string stability, which is hard to achieve in short headways for human driver or ACC, has become a design requirement (van Nunen et al., 2012) and also a basis for the various benefits of CACC. Previous studies have indicated that CACC can significantly improve roadway capacity (Liu et al., 2018), traffic flow quality (Van Arem et al., 2006) and energy efficiency (Shladover, 2018).
In despite of the great potentials, the usability of CACC may be seriously limited in the near future when the CAVs are mixed with a large number of non-CAVs. Although governments and manufacturers worldwide are making efforts to promote vehicular connectivity (Masini et al., 2018), CAVs are not likely to gain a dominant Market Penetration Rate (MPR) in the next two decades (Bansal and Kockelman, 2017). Consequently, the chance is rare to make CAVs travel consecutively in the mixed traffic. As most of the existing CACC systems require the connection with the nearest preceding vehicle (Naus et al., 2010, Milanes et al., 2014, van Nunen et al., 2012, Rajamani and Shladover, 2001, Ploeg et al., 2011), the CACC vehicle would have to degrade into ACC mode when its preceding vehicle is an unconnected vehicle.
There have been a few research efforts trying to address this issue of CACC. Graceful degradation of CACC (dCACC) (Ploeg et al., 2015) proposed to estimate preceding vehicle’s acceleration using onboard radar if it is not obtainable via communication. It was proven that with proper tuning of the state estimator (Kalman filter), dCACC can fulfill string stability at a short time gap which is less than a half of that needed for ACC. Nevertheless, it turned out that the noise in the radar-measured acceleration could compromise the smoothness of vehicle trajectory and the ride comfort (Ploeg et al., 2015). A class of methods called Connected Cruise Control (CCC) (Ge and Orosz, 2014, Zhang and Orosz, 2016, Ge and Orosz, 2018) explored the benefits of the communication with remote preceding vehicle when the nearest preceding vehicle is unconnected, but a limitation was that CCC considered the behavior patterns of unconnected vehicles to be known (or identified) and unchanged given different drivers. These assumptions would be challenged in the real traffic situation.
A CACC extension considering Unconnected vehicle in the loop, dubbed as CACCu, was recently proposed in (Chen and Park, 2020). CACCu enables a vehicle to follow an unconnected vehicle with string stability, by utilizing the communication with a further connected preceding vehicle. Different from the existing CCC, CACCu aims to be robustly string-stable given various unconnected vehicles’ car-following behaviors, without requiring identification process or extra information on the unconnected vehicles’ behavior. Numerical simulations with real traffic data showed that the proposed CACCu can improve string stability, control accuracy and fuel efficiency compared to existing methods (Chen and Park, 2020).
To validate the effectiveness and feasibility of implementation, the CACCu system must be developed on actual vehicles and tested in the field. The sensor error and nonlinear vehicle dynamics have been shown the two major factors that could significantly deviate the actual performance of automated driving from that in the simulations. On one hand, the unexpected sensor noise in the real traffic may undermine the smoothness of vehicle response and force the designer to adopt more conservative control parameters. Taking dCACC (Ploeg et al., 2015) as example, the vehicle trajectories in the experiments with actual radar were much jerkier than those in simulations assuming radar distance/speed errors in normal distributions. Thus, a “slower” state estimation and longer desired headway should be applied for the ideal ride comfort. Similarly, although it is proved that ACC with Constant Time Gap (CTG) policy can guarantee string stability as the feedback gains are sufficiently high (Xiao and Gao, 2011), none of the commercial ACC systems (Milanes et al., 2014, Gunter et al., 2020) were string-stable as they were not able to use such high gains under the constraints of sensor noise and ride comfort. On the other hand, the nonlinear dynamics of real vehicles (especially those with internal-combustion engines) could prevent the vehicles from achieving the acceleration/speed as commanded. Previous tests (Ge et al., 2018, Nieuwenhuijze et al., 2012) have shown that the accuracy of vehicle longitudinal control can be heavily affected by the limited power of engine, the power drops during gear shifts, and other uncompensated nonlinear behaviors of the vehicle. In summary, field evaluation with physical vehicles is always a necessary step to verify that an automated driving application can achieve its benefits in the real world.
Additionally, it is noted that many previous demonstrations of CACC or related applications shared a drawback in the experiment setting. Their test scenarios were commonly constructed with fabricated traffic disturbance, instead of the real-world situations. For example, the leading vehicle of the CACC platoon followed a trapezoid speed profile in (Milanes et al., 2014), triangle speed profile in (Ploeg et al., 2011, Wei et al., 2018), and step speed profile in (Ploeg et al., 2015). The Grand Cooperative Driving Challenge (GCDC) adopted a speed profile which consisted of three swept sines with frequencies 0.01 rad/s–2 rad/s (van Nunen et al., 2012). While these test scenarios could conveniently examine the desired properties (e.g., string stability) of the tested control methods, they could not give quantitative insight on how much benefit the proposed methods could achieved over the existing ones (e.g., ACC) in the real life. To reasonably quantify the performance improvements, more realistic test scenarios should be utilized in the evaluations.
In this paper, the CACCu is tested with CAVs in real-world-based driving scenarios. While the original CACCu algorithm (Chen and Park, 2020) determines the optimal acceleration of the vehicle, the control algorithm is re-developed in this study based on the experiment vehicle which only accepts speed commands. Then, the effectiveness of CACCu, in comparison with ACC and human driving, is evaluated in the car-following scenarios constructed using the NGSIM (Alexiadis et al., 2004) real-traffic data. Because the test scenario of CACCu involves the unconnected preceding vehicle which is human-driven, another challenge is that the human always drives differently in the multiple runs (Ge et al., 2018). The inconsistent human behaviors may undermine the fairness of the performance comparisons between different control methods. A more effective way, as shown later in this paper, is to replace the actual human-driven vehicle with an automated vehicle that is programmed to replay the recorded trajectory of an actual human-driven vehicle.
The rest of the paper is organized as follows: the second section introduces the automated vehicles used in the experiments. The third section describes the control design and parameterization of CACCu. The evaluation settings and the results are presented in fourth section. The key findings and concluding remarks are provided in the fifth section.
Section snippets
Experimental vehicles
Two experimental vehicles, Hyundai i30 PD and Hyundai Veloster, are shown in Fig. 1. The i30 PD plays as the ego vehicle ruled by CACCu, while the Veloster is used to create the test scenarios, as explained later in the Section 4.1. Both vehicles have been equipped with onboard sensors and throttle/brake pedal actuators. The V2V communications between vehicle are enabled by Wi-Fi modules. The transmission rate of Wi-Fi message is set 10 Hz to mimic the well-known Basic Safety Massage (BSM) of
Framework
It is a common strategy for a CACC vehicle to drive in ACC mode when encountering an unconnected preceding vehicle (Liu et al., 2018, Xiao et al., 2018). With CACCu, however, if there exists a further preceding vehicle which is connected, the ego vehicle may obtain additional benefits from the V2V communications, rather than completely falling back to ACC. For the ease of real-time implementation, this study pursues a linear time-invariant control design. Fig. 4 shows the feedback-feedforward
Experiment settings
In the evaluation of CACCu, ACC and human driving serve as performance baselines. It is noted that an actual human driver, instead of any driver model or trajectory data was employed to perform the “human driving” in the field experiment.
To ensure fair comparisons among CACCu, ACC and human driving, it is required that two preceding vehicles must drive identically each time when testing different control methods. However, a difficulty is that the 1st preceding vehicle is supposed to be a
Conclusions and future work
CACC with Unconnected vehicle in the loop (CACCu) is a potential application to extend the usability of cooperative longitudinal control of Connected Automated Vehicles (CAVs) in the mixed traffic. To validate the feasibility of CACCu for future implementation, this study developed a CACCu system based on real vehicles and evaluate it in the field. A speed-command-based CACCu algorithm is designed according to the identified longitudinal dynamics of the experimental vehicle. The controller is
CRediT authorship contribution statement
Daegyu Lee: Validation, Data analysis, Writing - original draft, Writing - review & editing. Seungwook Lee: Validation, Data analysis, Writing - original draft, Writing - review & editing. Zheng Chen: Conceptualization, Methodology, Data analysis, Writing - original draft, Writing - review & editing. B. Brian Park: Conceptualization, Methodology, Supervision, Writing - original draft, Writing - review & editing. David Hyunchul Shim: Writing - original draft, Writing - review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
This research is in part supported by the National Science Foundation under Grant No. CMMI-2009342.
References (36)
- et al.
Forecasting Americans’ long-term adoption of connected and autonomous vehicle technologies
Transp. Res. Part A Policy Pract.
(2017) - et al.
Adaptive model predictive control for a class of constrained linear systems based on the comparison model
Automatica
(2007) - et al.
Dynamics of connected vehicle systems with delayed acceleration feedback
Transp. Res. Part C Emerg. Technol.
(2014) - et al.
Experimental validation of connected automated vehicle design among human-driven vehicles
Transp. Res. Part C
(2018) - et al.
Modeling impacts of Cooperative Adaptive Cruise Control on mixed traffic flow in multi-lane freeway facilities
Transp. Res. Part C Emerg. Technol.
(2018) - et al.
Experimental comparative study of autonomous and co-operative vehicle-follower control systems
Transp. Res. Part C Emerg. Technol.
(2001) - et al.
Virginia tech comprehensive power-based fuel consumption model: model development and testing
Transp. Res. Part D
(2011) Connected and automated vehicle systems: Introduction and overview
J. Intell. Transp. Syst. Technol. Planning, Oper.
(2018)- et al.
Unravelling effects of cooperative adaptive cruise control deactivation on traffic flow characteristics at merging bottlenecks
Transp. Res. Part C Emerg. Technol.
(2018) - et al.
The next generation simulation program
ITE J. (Institute Transp. Eng.)
(2004)
Cooperative Adaptive Cruise Control With Unconnected Vehicle in the Loop
IEEE Trans. Intell. Transp. Syst.
Optimal control of connected vehicle systems with communication delay and driver reaction time
IEEE Trans. Intell. Transp. Syst.
Data-driven parameter estimation for optimal connected cruise control
Millimeter-wave technology for automotive radar sensors in the 77 GHz frequency band
IEEE Trans. Microw. Theory Tech.
Cited by (15)
Impact of beyond-line-of-sight connectivity on the capacity and stability of mixed traffic flow: An analytical and numerical investigation
2024, Physica A: Statistical Mechanics and its ApplicationsCollision Avoidance Motion Planning for Connected and Automated Vehicle Platoon Merging and Splitting With a Hybrid Automaton Architecture
2024, IEEE Transactions on Intelligent Transportation SystemsDistributionally Robust Optimization Based Model Predictive Control for Stochastic Mixed Traffic Flow
2024, IEEE Transactions on Intelligent Transportation SystemsA General Hierarchical Control System to Model ACC Systems: An Empirical Study
2024, IEEE Transactions on Intelligent Transportation SystemsAn AI-Assisted Systematic Literature Review of the Impact of Vehicle Automation on Energy Consumption
2023, IEEE Transactions on Intelligent VehiclesImpact of Platoon Management on Heterogeneous Traffic under V2X Communication Limitation
2023, Transportation Research Record