Intelligent bots for MOUT

• Main Author: Ting, Shang Ping
• Other Authors: Michael Harold Lees
• Format: Theses and Dissertations
• Language: English
• Published: 2011
• Subjects: DRNTU::Engineering::Computer science and engineering::Computer applications::Computers in other systems
• Online Access: https://hdl.handle.net/10356/43951
• Institution: Nanyang Technological University

Military Operations on Urban Terrain (MOUT) are defined as military actions that are planned and conducted on a terrain where man-made construction affects the tactical options available to the commander. This type of conflicts is characterized by street-by-street, room-by-room fighting. Strategies employed in MOUT differ immensely from fighting on other types of terrain, such as large deserts or jungles. The importance of MOUT warfare is increasing due to rapid world-wide urbanisation. The limitation of urban terrains means that soldiers are expected to function in small teams without the support of air-power or tanks. To survive, soldiers rely on small squad tactics and individual situational awareness. Modern armies may suffer substantial casualties in MOUT warfare. Going into MOUT operations without proper preparation could be very dangerous to the soldiers. Extensive training conducted with computer simulations may be able to reduce casualties in MOUT warfare. Simulations can introduce the trainees to the combat stress and shorten response time during actual combat. Although there had been considerable efforts and capital invested into this area, there are still many challenges facing existing simulation systems. One of the key challenges lies in the lack of realistic human behaviours. Human-like realism can improve the training relevance of MOUT simulations and allow the trainees to gain significant operational benefits. Human characters are represented by bots during MOUT simulations. It is important to develop intelligent bots in order to generate human-like behaviours. Our philosophy is that an intelligent bot should not only be able to generate seemingly realistic behaviors in some given situations, it should also behave like a human in the sense that the decision-making and cognitive processes of a bot should be similar to that of a human being. Therefore, our approach for human-like behaviours focuses on achieving both procedural and end-result realism for intelligent bot behaviours. We believe that by imitating human decision making and cognitive processes under various tactical situations, human-like bot behaviours can finally be generated. After interviewing soldiers from the Singapore Armed Forces (SAF) and engineers from the Defence Science and Technology Agency (DSTA), it is suggested that certain behaviourial skills are extremely important for human soldiers during MOUT warfare. These key skills are decision making, navigation and situation awareness. If the key areas can be modelled accurately, then realistic behaviours will be generated. To generate human-like behaviours, we study the decision making and cognitive process of the soldiers in the identified key skills. Subsequently, novel features are developed to improve the realism of bot behaviours in MOUT simulation. The novel features designed and implemented in this work includes a) a time critical decision making framework, b)a computational model for situation awareness and c) an autonomous navigation system. This report summarizes our current work on these novel features. They are designed to support our concept of closely imitiating the behaviours of humans to produce intelligent bot behaviours. We developed a time critical decision making framework for MOUT simulations called SNAP. As MOUT scenarios are generally time critical, the SNAP framework will be useful for generating realistic decisions during MOUT simulations. SNAP generates decision from incomplete information gathered under time constraints. The information comes from the situation awareness of the bots. A realistic computational model for situation awareness will greatly improve the quality of the information provided to SNAP. Therefore, we produced a computational model of situation