COOPERATIVE SEARCH BY MOBILE AGENTS UNDER UNCERTAINTY
Barouch Matzliach, Ph.D. candidate Advisors: Prof. Irad Ben-Gal & Dr. Evgeny Kagan
Via Zoom click here
The research addresses the problem of detecting multiple targets (static and mobile) by a group of mobile agents performing under uncertainty. It is assumed that the agents are equipped with sensors with positive and non-negligible probabilities of detecting the targets at different distances. The goal is to define the trajectories of the agents that can lead to the detection of the targets in minimal time. In contrast to the existing algorithms, it is assumed that the search is conducted under statistical errors of the first and the second types. Namely, both the probability that the agent does not recognize an existing target (false negative) and the probability that the agent recognizes a non-existing target (false positive) are greater than zero. The suggested solutions follow the classical Koopman's approach (1946) to an occupancy grid, while the decision-making and control schemes are conducted based on information-theoretic criteria. Sensor fusion in each agent and over the agents is implemented by using a general Bayesian update scheme. In the research, we have developed two types of detection procedures: the online reactive algorithms that implement the "expected information gain" approach utilizing the "center of view" and the "center of gravity" algorithms and a procedure based on a deep Q-learning scheme. The deep Q-learning algorithms maximize the cumulative information gain about the targets' locations and minimize the trajectory length on the map with a predefined detection probability. The deep Q-learning process utilizes a neural network technique that receives the agent location and current probability map and results in a preferred move of the agent. In the search for multiple targets by a group of mobile agents, the reactive Distributed Expected Information Gain (DEIG) algorithm implements dynamic Voronoi partitioning of the search space. It plans the paths by maximizing the expected one-step look-ahead information per region, and a Collective Q-max (CQM) learning algorithm finds the shortest paths of the agents by maximizing the cumulative information about the targets' locations using deep Q-learning techniques. Simulation analysis of the suggested procedures demonstrates their convergence to effective solutions even in scenarios where known existing algorithms are inapplicable. Also, comparing the existing methods in the appropriate scenarios demonstrated that the suggested algorithms, especially the Collective Q-max algorithm, considerably outperform the known methods. In particular, the proposed algorithms improve the results by 20% to 100% under different scenarios of noisy environments and sensors' sensitivity
Barouch Matzliach is a Ph.D. candidate in the Department of Industrial Engineering at Tel-Aviv University, working in the LAMBDA research group under the supervision of Professor Irad Ben-Gal. He holds a B.Sc. in Mechanical Engineering and an M.Sc. in Industrial Engineering, both from Tel Aviv University. Matzliach served in the IDF for over 30 years until 2018, specializing in developing and manufacturing land combat systems. He served as the Chief Engineer of the Merkava tank and later as Brigadier General, heading the Merkava & Armored Vehicles program in the Ministry of Defense. Currently, Matzliach provides consulting services for defense companies, focusing on the development and project management of military systems.