Modeling Reservation-based Autonomous Intersection Control in VISSIM

1 Autonomous vehicles are attracting more and more attention as a promising approach to improve 2 both highway safety and efficiency. Most previous studies on autonomous intersection 3 management relied heavily on custom-built simulation tools to implement and evaluate their 4 control algorithms. The use of the non-standard simulation platforms makes comparison between 5 different systems almost impossible. Additionally, without support from standard simulation 6 platforms, reliable and trustworthy simulation results are hard to obtain. In this context, this 7 paper explores a way to model autonomous intersections using VISSIM, a standard microscopic 8 simulation platform. Specifically, a reservation-based intersection control system, named 9 Autonomous Control of Urban TrAffic (ACUTA), was introduced and implemented in VISSIM 10 using VISSIM’s External Driver Model. The operational and safety performances of ACUTA 11 were evaluated using the easy-to-use evaluation tools of VISSIM. Compared with the optimized 12 signalized control, significantly reduced delays were resulted from ACUTA along with a higher 13 intersection capacity and lower volume-to-capacity (v/c) ratios under various traffic demand 14 conditions. The safety performance of ACUTA was evaluated using the Surrogate Safety 15 Measure Model, and presented few conflicts among vehicles within the intersection. Moreover, 16 the key steps and elements for implementing ACUTA in VISSIM are introduced in the paper, 17 which can be useful for other researchers and practitioners in implementing their autonomous 18 intersection control algorithms in a standard simulation platform. By using a standard simulation 19 platform, performance of different autonomous intersection control algorithms can be eventually 20 compared. 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 TRB 2013 Annual Meeting Paper revised from original submittal. Li, Chitturi, Zheng, Bill and Noyce 2 INTRODUCTION 40 With the rapid advances in sensing, information processing, machine learning, control theory and 41 automotive technology, wide application of autonomous vehicles on highway systems is no 42 longer a dream, but a reality in near future. Autonomous vehicles are vehicles without human 43 intervention (in-vehicle or remote) and are capable of driving in real-world highway systems by 44 performing complex tasks such as merging, weaving, and driving through intersections. Many 45 automotive manufacturers including General Motors, Ford, Mercedes-Benz, Volkswagen, Audi, 46 BMW, Volvo, and Cadillac have already begun testing their autonomous vehicle on highway 47 systems (1). Google is also developing and testing its Google driverless car. As of 2012, Florida, 48 Hawaii, Nevada, Oklahoma, and California have legalized or are considering legalization of 49 autonomous cars (1). All these facts indicate that the autonomous vehicles are set to appear on 50 road in near future. 51 Most field tests for autonomous vehicles were restricted to highway segment testing. 52 Intersection control of autonomous vehicles has been studied by researchers (2-19), however, 53 implementation in practice is difficult because intersections create more conflict points than 54 highway segments. For example, when vehicles arrive at an intersection from different 55 approaches, the right of way for traversing the intersection needs to be determined. Traditional 56 intersections use traffic control devices, such as stop signs and traffic signals, to regulate vehicles 57 right of way. For managing autonomous vehicles at intersections, the right of way may be 58 controlled by an intersection central controller through vehicle-infrastructure (V2I) 59 communications (2-12), or through negotiation between vehicles via vehicle-vehicle (V2V) 60 communications (13-17). 61 Studies have been conducted to explore ideas and algorithms for managing autonomous 62 vehicles at intersections. By control strategy, the autonomous intersection control can be 63 classified into centralized control and decentralized control. For centralized control, all vehicles 64 establish communication connections to an intersection central controller, or intersection 65 manager (2-12). The intersection manager determines the vehicles’ passing sequence. In a 66 decentralized control system there is no intersection manager. The passing sequence is typically 67 negotiated by vehicles based on a certain protocol (13-17). Among all these available solutions, 68 the reservation-based centralized control system has been found to work best for urban 69 intersections with high traffic demand because of its mechanism of maximizing the intersection 70 capacity (14). 71 Due to the complexity of field implementation, most researchers used traffic simulation 72 to validate their developed strategies for autonomous intersection control. However, none of the 73 exiting studies used standard commercial traffic simulation software such as VISSIM or 74 CORSIM when evaluating the performance of their proposed strategies. Rather, simulation tools 75 developed by the respective authors were used in the evaluation process, which made the results 76 less reliable and hard compare with each other. In addition, it was noticed that most existing 77 studies lacked standard usage of terms and clear description of simulation parameter settings 78 when presenting the evaluation results. For example, when presenting the traffic volume, no 79 clarification of whether the volume is per lane or per entire approach was presented. Also, terms 80 to define lane configurations, speed distribution, volume, and delay, as well as the number of 81 runs per experiment, random seed selection, and simulation period were excluded from the 82 analyses, or were not consistently defined across different studies. Most likely the inconsistency 83 is due to the usage of different custom-built simulation software programs, rather than standard 84 commercial simulation software packages. 85 TRB 2013 Annual Meeting Paper revised from original submittal. Li, Chitturi, Zheng, Bill and Noyce 3 Standard simulation packages like VISSIM and CORSIM can provide standard parameter 86 settings and outputs. In addition, using the standard package can guarantee reliable vehicle 87 generation, car-following, lane-changing, and many other driving behavior related modeling in 88 the simulation. Flexible settings of speed distribution, heavy vehicle percentage, and 89 distributions of acceleration and deceleration rates can also be simply achieved, along with 90 strong evaluation outputs like travel time and delay. Moreover, commercial packages like 91 VISSIM have options to output vehicle trajectories, which can be directly imported into 92 Surrogate Safety Assessment Model (SSAM) to analyze the safety performance of the 93 intersection (19). 94 Wu et al. indicated in their paper that they chose to develop their own simulation tool 95 rather than use standard traffic simulation packages such as VISSIM, AIMSUN, or PARAMICS, 96 because the standard packages do not allow vehicles to be controlled individually (14). In fact, 97 VISSIM offers flexible customization functions to facilitate building different special 98 applications through APIs and COM extensions. All these functions offer the potential to 99 implement applications for autonomous intersection control. In this paper, implementation of a 100 reservation-based system in VISSIM using VISSIM’s External Driver Model is presented. The 101 establishment of the simulation model, implementation of the reservation-based control 102 algorithm, and finally evaluations of operational and safety performance are discussed. 103 104 ENHANCED RESERVATION-BASED AUTONOMOUS INTERSECTION CONTROL 105 A reservation-based system utilizes a centralized control strategy for managing fully-autonomous 106 vehicles at an intersection. All vehicles in a reservation-based system communicate only to a 107 centralized intersection controller, namely, intersection manager (IM). The IM regulates the 108 intersection by determining the passing sequence of all the approaching vehicles (2-10). 109 110 111 112 FIGURE 1 Intersection mesh of tiles and example of vehicle’s possible routing decisions. 113 114 The system presented in this paper is named as Autonomous Control of Urban TrAffic 115 (ACUTA), which is developed based on First-Come-First-Serve (FCFS) reservation-based 116 protocol (2) with enhancements to improve some operational issues identified in previous studies 117 (2, 8). These issues include the “starvation” issue where approaching vehicles on the side street 118 N TRB 2013 Annual Meeting Paper revised from original submittal. Li, Chitturi, Zheng, Bill and Noyce 4 cannot get reservations when the traffic demands on the major and side street are unbalanced; 119 and (2) slow-speed reservation issue which unnecessarily occupies many intersection resources. 120 ACUTA regulates an intersection which is divided into a mesh of n by n tiles, as shown in Figure 121 1, where n is termed as granularity, and reflects the tile density of the intersection mesh. 122 In ACUTA , each approaching vehicle sets up a communication connection with the IM 123 after it enters the IM’s communication range. When connected, the vehicle immediately sends 124 the IM a reservation request along with the vehicle’s location, speed and routing information (i.e., 125 making a left/right turn or going straight), indicating its intention to traverse the intersection. The 126 IM processes the reservation request by computing the required time-spaces for the vehicle to get 127 through the intersection (i.e., intersection tiles that will be occupied by the requesting vehicle for 128 all simulation steps when the vehicle traverses the intersection) based on the location, speed, 129 maximum acceleration rate, and the routing information provided by the requesting vehicle. 130 Acceleration from the requesting vehicle’s current location to the entrance boundary of the 131 intersection is considered when computing the required time-spaces. Using different acceleration 132 rates can change the required time-spaces significantly. The alternative acceleration rate shall fall 133 within the range from zero to the maximum acceleration rate of the particular vehicle, and is 134 calculated using the following equation. 135 max max 0 ( 1) 1 ( 1) ( 1) i