Using a Fleet of Low-Cost Ground Robotic Vehicles to Play Complex Games: Development of an Artificial Intelligence (AI) Vehicle Fleet Coordination Engine GOALS. The proposed research shall focus on developing an artificial intelligence (AI) engine for a fleet of low-cost multi-capability ground robotic vehicles. More specifically, the main goal of the project is to develop, test and demonstrate an AI software gaming engine that permits vehicles to play simple and complex games; e.g. Tic-Tac-Toe [1], Connect 4 [2], a Chess-like game which involves 5 main pieces and two pawns (7 robots on each side) [3]-[6], and a more complicated GO-like game which requires surrounding an opponent s forces (again 7 robots on each side) [7]-[8]. MOTIVATION. The above is a very worthwhile activity because it combines precision control of mobile ground robots with complex algorithms [9]-[12]. It is complex algorithms when combined with sensors and actuation devices for real-time decision making which results in intelligent systems. Collectively, the two will represent a powerful test bed for the deployment and visualization of complex algorithms. Such an AI-enhanced multi-vehicle fleet can be used in each of the following important applications: (1) performance or objective-based coordination of multiple vehicles, (2) coordination of military robots/forces, (3) coordination of first responder robots/resources/agents, and (4) using the AIenhanced fleet to entertain prospective elementary school, middle school, high school and community college students that are thinking about engineering as a career; i.e. the fleet would be a great outreach tool. Broader Impact. Dr. Rodriguez plans on using the AI-enhanced fleet as engineering recruitment tool as he pursues his Academic Success and Professional Development (ASAP) Project-Based Engineering Academy Across Arizona campaign a venture involving over 15 community colleges across the state (as well as their local high schools). OBJECTIVES. The paramount objective of the proposed work is to develop several look-ahead performance-based algorithms that can be used to play the games mentioned above; i.e. Tic-Tac-Toe [1], Connect 4 [2], a Chess-like game [3]-[6], and a more complicated GO-like game [7]-[8]. CRITICAL QUESTIONS. Critical questions to be addressed in this work are as follows: (1) Look-Ahead Algorithms. What look-ahead algorithms should be implemented [9]-[12]? We do not want to consider more than 2-3 algorithms. 2-3 algorithms will be enough to systematically compare the performance of each. (2) Opponent Fleet Look-Ahead Capability. How does the performance of the algorithms depend on the look-ahead capability level of the opponent fleet? As the look-ahead capability of the opponent fleet improves, we expect algorithm performance to degrade. This shall be quantified by conducting specific empirical (simulation) studies. (3) Time-To-Solution. How does the time to compute a solution (move) depend on the look-ahead aggressiveness (i.e. the number of moves that we look ahead)? We expect the algorithms to be
non-polynomial time algorithms. That is, the time to compute a solution grows exponentially (non-polynomial) as the number of possible moves grows. (4) Algorithms that Improve Over Time. How do we get the algorithms to improve over time [6]; i.e. get better as they are exercised and acquire new knowledge? Many methods for doing this have been examined in the literature. We shall explore a few. (5) Robot Spatial Command Following. How do we issue spatial commands to the robots in the fleet? The answer here is provided within the following very detailed (377 page) MS thesis that Dr. Rodriguez supervised [13]. A simple vision-based Cartesian stabilization control system will do the job [14]. Establishing a Foundation for Future Work. In this work, it is the algorithms that represent the critical catalyst for the effort. While the control is essential so that the robots can move to their solutioncomputed destinations on the playing plane, it has been figured out and well documented within the aforementioned MS thesis [13]. Given this, it is natural to ask: Why bother with the controls portion of this project? The answer here is twofold. First, watching robots play chess will be exciting for many. Secondly, and more fundamentally, we view the proposed research as representing a starting point for the development of much more sophisticated fleet coordination. This could involve more complex games; e.g. soccer, basketball. At some point, we will advance to total warfare between fleets (physical and cyber). Given this, the potential impact of the proposed project can be very significant particularly when one factors the Arizona-wide impact that Dr. Rodriguez Engineering Academy efforts will have on students across Arizona; particularly low-income students, woman and underrepresented minorities. Final Demonstration. The final demonstration will involve robots playing Tic-Tac-Toe [1], Connect 4 [2], 7 piece Chess [3]-[6] and Go [7]-[8]. All results will be documented in a final comprehensive report and video. The report shall be used as the basis for submitting our work for publication in a refereed journal; e.g. IEEE Transactions on Control Applications.
References [1] J. Grim, P. Somol and P. Pudil, Probabilistic neural network playing and learning Tic-Tac-Toe, Pattern Recognition Letters, vol. 26, no. 12, pp. 1866-1873, 2005. [2] A. Verma, Genetic evolution of connect four strategies, ProQuest Dissertations and Theses, pp. 198, 2008. [3] C. Ewerhart, Backward Induction and the Game-Theoretic Analysis of Chess, Games and Economic Behavior, vol. 39, no. 2, pp. 206-214, 2002. [4] S. Bushinksky. Deus ex machina-a higher creative species in the game of chess. AI Magazine 30(3), pp. 63-70. 2009. [5] T. Shunhua and C. Miao, Search Algorithm in Five-Piece Chesss, Journal of Computing, vol. 3, no. 4, pp. 596-599, 2012. [6] D. Fogel, T. Hays, S. Hahn and J. Quon, A self-learning evolutionary chess program, Proceedings of the IEEE, vol. 92, no. 12, pp. 1947-1954, 2004. [7] K. Chen, Maximizing the chance of winning in searching Go game trees, Information Sciences, vol. 175, no. 4, pp. 273-283, 2005. [8] J. Hoock, C. Lee, A. Rimmel, F. Teytaud, M. Wang and O. Teytaud, Intelligent Agents for the Game of Go, IEEE Comput. Intell. Mag., 2010. [9] K. Hausken and G. Levitin, Minmax defense strategy for complex multi-state systems, Reliability Engineering & System Safety, vol. 94, no. 2, pp. 577-587, 2009. [10] F. Facchinei, J. Pang and G. Scutari, Non-cooperative games with minmax objectives, Computational Optimization and Applications, vol. 59, no. 1-2, pp. 85-112, 2014. [11] K. Chellapilla and D. Fogel, Evolution, neural networks, games, and intelligence, Proceedings of the IEEE, vol. 87, no. 9, pp. 1471-1496, 1999. [12] J. Mandziuk, Towards Cognitively Plausible Game Playing Systems, IEEE Comput. Intell. Mag., vol. 6, no. 2, pp. 38-51, 2011. [13] Z. Lin, Modeling, Design and Control of Multiple Low-Cost Robotic Ground Vehicles, ASU MS Thesis, Electrical Engineering, (Supervisor: Dr. A.A. Rodriguez), 377 pages, August, 2015. [14] F. C. Vieira, A. A. D. Medeiros, P. J. Alsina, A.P. Araujo, Position and Orientation Control of a Two-Wheeled Differentially Driven Nonholonomic Mobile Robot, ICINCO Proceedings, 7 pages, 2004.
TIMELINE FOR Using a Fleet of Low-Cost Ground Robotic Vehicles to Play Complex Games: Development of an Artificial Intelligence (AI) Vehicle Fleet Coordination Engine Semester: Spring ZZZ Comprehensive Literature Survey October 2015-Jan 2016 Build 2 Robots January 2016 Tic-Tac-Toe and Connect 4 January- February 2016 Chess February-March 2016 GO March-April 2016 Data Collection, Add Learning Capability April 2016 Final Data Collection, Demonstration, Final Report April-May 2016
BUDGET FOR Using a Fleet of Low-Cost Ground Robotic Vehicles to Play Complex Games: Development of an Artificial Intelligence (AI) Vehicle Fleet Coordination Engine Semester: Spring ZZZ 2 Enhanced Thunder Tumbler Robotic Vehicles $315.32 Books, Supplies $84.68 TOTAL: $400 Cost Breakdown for Enhanced Thunder Tumbler Robot Product/Component Quantity Price ($) $ Thunder Tumbler Vehicle 1 $10 $ Raspberry Pi 2 Model B 1 $40 $ Arduino Uno 1 $ $12.19 Adafruit Motor Shield 1 $ $22.50 Raspberry Pi 5MP Camera 1 $25 $ Camera Holder 1 $ $5 HCSR04 Ultrasonic Sensor 1 $ $1.87 Power Supply for Raspberry Pi 1 $10 $ Power Supply for Arduino 4 $ $6.75 Magnetic Wheel Encoders 2 $ $4.40 Adafruit 9dof IMU 1 $ $19.95 Total Price $ $157.66 Two Enhanced Thunder Tumblers will be built so that we can progress toward a fleet size of approximately 16 vehicles.