VEHICULAR 2020 : The Ninth International Conference on Advances in Vehicular Systems, Technologies and Applications A Multi-Agent Approach to Simulate Autonomous Traffic with Games: How to Transform GTA-SA/SA-MP in Your Simulation Platform Jonataˆ N. Cirqueira∗, Pedro C. Mesquitay, Rodrigo R. Novaes Jr.z and Sandro R. Diasx Centro Federal de Educac¸ao˜ Tecnologica´ de Minas Gerais Belo Horizonte, Brazil E-mail: fjonatanc0511∗, pedrocesar1410y, [email protected], [email protected] Abstract—Urban mobility is among the main problems of the network composed by 5% of self-driving cars already demon- contemporary society. In this context, the vehicle automation strates significant advantages in the traffic flow [4]. technologies stand out in several aspects, such as reduction in Many simulation platforms were built to test autonomous accidents, congestion and emission of pollutants. Considering traffic problems and solutions. Having a variety of them is that, this work seeks to upgrade the simulation platform com- important to validate different scenarios and perspectives. The posed by Grand Theft Auto: San Andreas, and its modification San Andreas: Multiplayer (GTA-SA/SA-MP) that, according to two main perspectives for autonomous vehicles simulations the literature, is an appropriate environment for implementing are two-dimensional (2D) and three-dimensional (3D). The autonomous vehicles networks. By using available structures in 3D perspectives are closer to reality, containing more details this environment, it is possible to perform three-dimensional and problems to solve, like relief dynamics, consumption simulations in many scales, ranging from a single village to cost when driving uphills and downhills, among others [5]. an extensive intercity map. From the available structures in However, due to the complexities and costs in building 3D game, it was used a set of navigational nodes, which identify software, most of the simulations are 2D, like the Texas all the available routes for the vehicles in the roads. We used University’s AIM project [6][7]. the navigational nodes to generate a weighted directed graph, Considering that, games like Grand Theft Auto: San An- to which we applied Dijkstra’s and A* search algorithms. From dreas (GTA-SA) [8] and its modification San Andreas Multi- that, it was observed that the vehicles were able to calculate and follow the best route from a source to a target node in any place player (SA-MP) [9] team up to form a good environment to of the map, obtaining a realistic environment to simulate and simulate multi-agent systems. This platform leverages many test solutions for the traffic, such as the autonomous intersection of the game’s features, like an extensive 3D map containing management protocol, which will be implemented in future. multiple cities, physical models for mechanics and collision, Keywords–Autonomous vehicles; multi-agent system; game; different vehicle models, weather manipulation and many VANET; IoT. others [10]. Previous works have managed to successfully build an I. INTRODUCTION autonomous vehicles network inside GTA-SA/SA-MP [10]. Urban mobility has proved to be one of the main problems Some of them could even build autonomous intersection we face in society. According to Texas A&M Transportation management protocols, which is one of the hottest topics in Institute, in 2017 the delays of industrial deliveries, along with autonomous traffic problems [11]. However, these works were the fuel consumption, caused a congestion cost of US$ 179 expanded only to a small village in the game, being unable to billion in the 494 urban areas of United States. Moreover, scale all solutions for the whole game map [10]. Therefore, an average auto commuter wasted 21 gallons of fuel only at this work sought to implement an autonomous traffic system traffic congestion. The report also attested that average auto using the entire GTA-SA’s map, whose detailed goals can be commuters spent 71 hours of extra travel time in the same described as: year [1]. Besides monetary cost, many fatalities happen in the traffic. • Use the game’s structures to identify vehicle’s paths According to the World Bank Group, about 1.25 million people along the game map; die on world’s roads every year, while in average 20 or 50 • Apply routing algorithms to understand if the game million people are seriously injured. In America, traffic has provides a consistent environment for simulations. been the leading cause of death since 1975 [2]. We could adapt GTA-SA/SA-MP’s features to simulate In addition, traffic congestion is a major aggravating factor autonomous traffic in the whole game map. Also, we found for environmental problems. Vehicle emissions became the paths data for different objects, like pedestrians and boats, dominant source of air pollutants, raising the risks of morbidity opening possibilities to interact with human agents and to and mortality, specially for commuters and individuals who develop sensor networks for sea navigation. live near roadways. When traffic flow is slow, regular speedups In Section I, the reader had a vision of the existing and slowdowns can increase travel time and diminish the problems traffic automation want to solve, as well as some dispersion of pollutants, elating by four times the emission tools that can be used to achieve that. In Section II, we of CO, HC and NOx [3]. will present the base concepts and methodologies applicable. One of the emerging solutions for these problems is the Section III describes the implemented methodology in a way traffic automation, currently leaded by the development of to achieve the described goals. In Section IV, we expose the autonomous vehicles. They are proving to be effective and results and data acquired from this work. In Section V, we efficient, mainly due to their ability of taking deterministic do a critical analysis on the data, as well as briefly explore and accurate decisions. Researches indicate that a vehicular future works. Copyright (c) IARIA, 2020. ISBN: 978-1-61208-795-5 26 VEHICULAR 2020 : The Ninth International Conference on Advances in Vehicular Systems, Technologies and Applications II. CONCEPTS &METHODOLOGY fP1;:::;Pk−1;Pk;Pk+1;:::;Png have sub-paths from P1 to This section will use theoretical concepts to present meth- Pk−1 and Pk+1 to Pn which are minimal. However, there is a 0 ods to achieve our goals. In Section II-A, we do an analysis minimal path Pk whose distance is lesser than Pk. Therefore, 0 0 on how to use games to simulate an autonomous multi-agent the distance of P = fP1;:::;Pk−1;Pk;Pk+1;:::;Png is system; this will require software engineering and design pat- lesser than P , meaning that concatenating minimal sub-paths terns understanding. In Section II-B, we’ll understand which will result in a minimal path. math and computational models, as well as algorithms, are The process of identifying a path with minimal cost is necessary to apply in this system. In Section II-C, we define based on Lemma 1. Therefore, Figure 1 lists an algorithm the problems we want to solve with this model, as well as the to calculate all minimal paths from a single-source vertex of high-level solution that might fit in any kind of system. a graph. A. Games and design patterns 1: for all u 2 V (G) do We can use two object-oriented programming principles 2: d(u) 1 fInitial distance is unknowng to say that an electronic game or software product is eligi- 3: p(u) ; fIdentify which vertex is incident over u in ble to simulate autonomous agents and multi-agent systems: the minimal pathg open/close and the dependency inversion principles [12]. 4: end for Definition 1: Eligibility of a software product to be used 5: d(s) 0 in modifications. Let M be a software product, M can only be 6: INSERT HEAP(A, s) fCreates a priority queue A based extensible to modifications when the following characteristics on the distance to sg are present: 7: while jAj ≥ 1 do 1) M’s modules and libraries may be extended without 8: u REMOVE HEAP(A) fRemoves the vertex with being modified; minimal distance from Ag 2) There is a set of public interfaces of M such that they 9: for all uv 2 E(G); v 2 V (G) do may be implemented, incorporated and distributed 10: if d(v) > d(u)+ CALCULATE COST(uv) then with M. 11: d(v) d(u)+ CALCULATE COST(uv) By identifying an application that fits Definition 1, we will 12: p(v) fug reduce and decouple the amount of work. The responsibility 13: INSERT HEAP(A, v) of simulating autonomous agents and multi-agent systems can 14: end if be addressed to the researcher, while the dependencies of this 15: end for work, like physical models, climate, objects and events can 16: end while rely on the modified application only. 17: return (p; d) fReturns a pair with the minimal distances d to all vertexes and a set p that identifies incidencesg B. Routing & mechanics models Graph theory is the study of the relation between elements Figure 1. Algorithm to calculate the minimal cost of a path inside a graph of a set. G = (V; E) is said to be a graph, where V is a set of G = (V; E), such that the path starts from s 2 V (G) and terminates in all V vertexes and E = (u; v) is a set of pairs, where u; v 2 V (G), other reachable vertexes in . meaning that a vertex u is related to v. A graph is directed when there is a relation over all edges (u; v) 2 E(G) such The algorithm in Figure 1 can be implemented in two well- that u is said to be incident over v [13].