AD VANCED TRAFFIC ENGINEERING ASSIGNMENT CIV 8331
- 1. BAYERO UNIVERSITY KANO FACULTY OF ENGINEERING CIVIL ENGINEERING DEPARTMENT ADVANCED TRAFFIC ENGINEERING BY ABDULRASHID HASSAN MUSA Reg No: SPS/19/MCE/00027 ASSIGNMENT ON: REVIEW OF MICROSCOPIC TRAFFIC MODEL USING ARTIFICIAL INTELLIGENCE SUBMITTED TO PROF. H.M.ALHASSAN DECEMBER, 2022
- 2. Roadway safety continues to be a major public health concern. Recent statistics show that more than 30,000 fatalities occur due to motor vehicle accidents, and in the year 2012, motor vehicle crashes resulted in more than 2 million injuries. As a result of these ongoing trends, simulators continue to become more abundant in applications ranging from Intelligent Transportation Systems (ITS) research, autonomous driving, human factors studies, rehabilitation, and driver training and workload applications. However, many current commercial simulators lack realism with regards to accompanying traffic, which often does not satisfactorily respond to the real-time actions of the human subject who is operating the simulation. Artificial traffic simulation models found within commercial driving simulators are often “macroscopic” in nature – they aggregate the description of overall traffic flow, which is based on “idealistic” driver behavior. This lack of network realism (particularly in the vicinity of the human subject operating the simulator) limits the application scope. INTRODUCTION
- 3. In a simple term Artificial intelligence is the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings (Goggle search engine). While microscopic traffic flow models are a class of scientific models of vehicular traffic dynamics that describe and simulate single vehicle-driver units, like the position and velocity of a single vehicle. AI is in the spotlight as one of the emerging fields transforming the transport sector. It is not a new term. Academics ups and downs future progress talked about artificial intelligence as early as the 1950s. Since then, AI has undergone a number of, where optimistic expectations were followed by bitter disillusion. In recent years, AI has made a lot of progress, as machine learning techniques have been combined with technologies used for searching and analyzing the large quantities of data (otherwise known as big data and data mining) produced by the development of the digital world. Other reasons for its successful growth include the development of communications networks and the internet of things, as well as progress in transport devices. The AI in transport is expected to be even more spectacular, although there is no agreement on the timing and exact nature of these developments LITERATURE REVIEW
- 4. Applications of Artificial intelligence Self-driving car Machine learning Natural language processing Pattern recognition Algorithm Data security Robotic processing automation Etcetera
- 5. AIMS AND OBJECTIVES OF THE RESEARCH To get a deeper understanding of microscopic traffic models using artificial intelligence
- 6. DEFICIENCIES OF THE PREVIOUS RESEARCH IN THIS AREA Related research has noted that existing microscopic traffic simulation models (based on available car- following, gap-acceptance, and lane-changing models) often lack the level-of-detail required for safety evaluations, which demand models that reflect errors in drivers’ perception, decision-making, and actions (Cunto and Saccomanno, 2006). Other researchers (e.g., Punzo and Ciuffo, 2011) have emphasized the four main requirements for appropriately integrated (TS-DS) simulation models. These are: Accurate road matching between traffic and driving simulators; Synchronization of traffic and driving modules with real time; Consistency of the updating calculation frequency; and Management of autonomous vehicle visualization.
- 7. In particular response to these four stated research needs, and to the various shortcomings of the related work that came before ours, we have constructed our own Artificially Intelligent Transportation Model (AITM) intended for clinical, training, and research simulation applications. In this paper, we describe the design of the major components of the AITM, and some of its preliminary shortcomings that necessitated critical modifications. Prior to that discussion, we first present ideas for the extensibility of our work to related domains in M&S research. BROADER IMPACTS SIMULATION HARDWARE ENVIRONMENT DESCRIPTION
- 9. Figure 1-Driving Simulator (Interior)
- 10. Figure 2 – Driving Simulator (Exterior)
- 11. Figure 3- Virtual Driving Environment
- 12. AITM DESIGN (PRELIMINARY) METHODOLOGY In this section, the primary features of the AITM are described, including the motion model (both for linear motion and for turning motion), collision detection between moving objects (both between AI vehicles, and the human driven participant), models describing vehicle behavior both at signalized intersections and at stop signs, the AI vehicle lane-changing model. This section concludes with an overview of major shortcomings of the preliminary model that necessitated further development.
- 13. Determining if any two 3D objects intersect – a technique commonly known as collision detection - is not an easy task. The key to optimize these calculations is to efficiently discard non-colliding objects before applying a full collision test (to all other candidate objects in the scene). The typical path for discarding objects is to first divide a virtual scene into segments, while keeping track of which segment the object in question is located, and then using collision tests with all the non-discarded objects in that segment. One common procedure is known as the Oriented Bounding Box (Eberly, 2008). COLLISION DETENTION
- 14. The Oriented Bounding Box This method rotates the bounding box with the geometry so that the bounding box represents the geometry of the object, even when rotated. This technique was chosen for the current effort, as the AI vehicles traverse within specific lanes of traffic which are at different orientations, and collision will be detected with other vehicles within the same lane (or adjacent, parallel lanes of travel). Also, a bounding box serves as a sufficiently accurate representation of the shape of a motor vehicle.
- 15. As shown in Figure 4- above, the bounding box is attached to the AI vehicle, and new coordinates are calculated (using matrix transformations) each time the AI vehicle moves or rotates. In addition, the bounding box was extended in the direction of the heading of each AI vehicle to detect collisions with other moving vehicles. The length of the bounding box was assigned to be dependent on both travel speed and the posted speed limit Figure 4 – Oriented Bounding Box (3D)
- 16. TRAFFIC SIGNAL MODEL Now that the AI vehicles are able to detect (and avoid) collision with one another, the next step was to add logic for AI vehicles when in the vicinity of signalized intersections. This algorithm handles all decisions to be made depending on the current “status” (i.e., red (R), yellow (Y), green (G)) of each intersection. Obviously, the AI vehicles need to know the state of each signalized intersection as they traverse the virtual environment. A particular 3-bay (G-Y-R) traffic signal at a 4-way intersection has two sides, and therefore, a total of five possible basic states (i.e., R-R, G-R, G-Y, Y-G, R-G), discounting turning arrows, flashing lights, and other more advanced signal states.
- 17. Figure 5-Left Turn Detection Figure 6-Stop Sign Detection
- 18. STOP SIGN MODEL The scope of the traffic signal algorithm was expanded to include all traffic signals within the current virtual environment and its logic was extended for application at intersections governed by stop signs. The algorithm was designed to account for “all-way” stop signs and “single way” stop signs. For the former, the algorithm keeps track of the vehicles waiting at the stop sign, and reserves slots for each vehicle as soon as it reaches the stop sign. According to the slot reserved, each vehicle is given a “go” signal to cross the intersection. As soon as an AI vehicle starts moving through the intersection, a programmer indicator internally denotes that the intersection is occupied, and the algorithm waits for that vehicle to clear the intersection.
- 19. LANE CHANGING MODEL Lateral movement (e.g., lane changing) by AI vehicles is necessary to represent real environment traffic scenarios. For example, the AI vehicles in the left-most lane will take a left turn at an approaching intersection, vehicles in middle lanes will continue straight, and the vehicles in the right-most lane will turn right (or go straight). These decisions were made using stochastic effects, as each AI vehicle approaches each given intersection. For an AI vehicle to change lanes, the heading angle no longer conforms to the (global) heading angle with the street upon which the vehicle is traveling. To assure that the lane change appears fluid, the lane changing function increments this heading angle in small amounts (heuristically chosen to be approximately 2.5 degrees) at each frame rendering, as demonstrated in the figure below
- 20. Figure 7 – Lane Changing Transition Zone Figure 8 – Lane Changing Collision Detection
- 21. DEFICIENCIES OF THE CURRENT MODEL AND IMPROVEMENT Once the preliminary model was designed, we were able to perform in-house testing and validation to observe the operability of the AI Traffic Model. In so doing, a number of deficiencies were noted that were difficult to forecast during the initial development phase. Three such deficiencies are described and illustrated here with some detail, along with brief descriptions of how these model deficiencies were ultimately improved, or circumvented altogether. Perhaps by offering these challenges, we will provide insight for future application developers in transportation based M&S who might face similar difficulties.
- 22. Figure 9 – Odd-shaped intersection
- 23. Figure 10 – AI vehicles waiting to turn
- 24. Figure 11 – AI-Human Interactions
- 25. CONCLUSIONS Roadway safety and sustainability continue to be major public health concerns, and subsequently, simulators (and other M&S technologies) continue to become more abundant in a wide variety of Intelligent Transportation Systems (ITS) research applications (e.g., autonomous driving, human factors, and rehabilitation). To confront these problems, standalone simulators are often implemented as an analysis and decision-making tool. Driving Simulators (e.g., to monitor driver behavior, performance, and attention), and Traffic Simulators (e.g., to plan, design and operate transportation systems) have been employed with some success. However, while traffic simulation models allow for capturing dynamics of full-scale traffic networks, they often lack behavioral realism. Largely for these reasons, an Artificially Intelligent Traffic Model (AITM) was constructed to operate in conjunction with a custom-designed driving simulation environment. The framework presented in this paper has been designed as an alternative for commercial “microscopic” traffic simulators, whose operability is often more concerned with gross end mass vehicle behavior, with individual vehicle movements that are not fluid, and therefore not suitable for integration with live human subjects.
- 26. While vehicle and transportation network technologies continue to evolve, there is an increasing urgency for improved fidelity for driving/traffic simulation research. As such, the research described here can be expanded and improved in numerous ways, and to conclude the paper, a few detailed suggestions are offered below: RECOMMENDATION
- 27. Implement customized human behavior models to enhance the traffic mobility: The proposed concept is to override the (artificially intelligent) driver behavior model with experimentally attained, subject-specific human behavior performance models.
- 28. Provide a high-fidelity, multiple-participant capability to facilitate research that involves real-time interaction between human participants. Two or more driving simulators should be able to connect in real-time, which would enable human drivers to interact with each other. Despite the success of driving simulators such as DiVE (Prendinger et al., 2014) that have already provided certain level of multiple driver capacity, there is still room for improvement in terms of functionality and fidelity
- 29. Integrated Traffic-Driving-Network Simulation The microscopic traffic model presented in this work has been the first step towards the development of a more complete integrated simulation environment for transportation research. While there have been numerous studies that have attempted to develop integrated two-way simulators (e.g., traffic-driving, traffic-network), none has attempted to integrate all three types of simulators. To this end, a 3-in-1 Integrated Traffic-Driving-Network Simulator (ITDNS) is currently under development and ongoing enhancement (Zhao et al., 2013).
- 30. THANK YOU FOR LISTeNING