Opportunistic Multi-robot Environmental Sampling via Decentralized Markov Decision Processes

Energy-Efficient Blockchain-Enabled Multi-Robot Coordination for Information Gathering: Theory and Experiments

“…Mobile robots can be used to autonomously gather meaningful information based on which future actions can be taken. Due to its sheer practical significance, the domain of information sensing using autonomous mobile robots has recently received considerable attention [1,[8][9][10][11][12][13][14][15][16]. In this problem setup, the goal of the robot(s) is to plan paths of lengths k such that the maximum amount of information can be collected from the environment.…”

“…, where |Σ| denotes the covariance matrix's determinant, while |V| denotes the vertex set's cardinality. It is a standard assumption in kernel-based parameterizations of GPs that the correlation between two nodes is inversely proportional to the distances between them [4,10,21]. We exploit this property when computing entropy by approximating the computationally intensive matrix determinant |Σ| by the product of the per-node variances (σ 2 v ) along the diagonal of Σ.…”

“…Under these circumstances, opportunistic connectivity is the go-to option-in this, the robots are not required to maintain communication with others, but if two or more robots are within each other's communication ranges, they will form an ad-hoc network to share data as required. Opportunistic connectivity has shown to be effective in multi-robot tasks [9,10,32]. For a survey of available connectivity models and their applications in multi-robot systems, the reader is referred to [33].…”

“…where Cv 0 i denotes the length-|V| row vector of all zeros except for a one in component v 0 i , and Cv 0 i is its matrix transpose. During periodic or opportunistic connectivity, the posterior statistics will sometimes evolve on a batch of measurements, which is easily accommodated by appropriate augmentation of the output matrix C(•); the reader is referred to [10] for more details. As the sensing step concludes, and each robot now possesses its updated GP statistics GP 0 i = (µ 0 i , Σ 0 i ) via Equation ( 2), the per-node rewards using Equation (1) are calculated.…”

“…, 16. The access patterns (1,2,3,4) or (5,6,7,8) are examples of totally serial execution given the restriction in DDR that only one chunk may be placed inside a certain cache C i at any one time, whereas (1,5,9,13) or (3,8,10,13) are examples of completely parallel execution. Technically, according to a [61]-derived formula, the amount of energy used by an algorithm A with the execution time τ for a P bank DDR3 architecture with B bytes per chunk is provided by:…”