UAV Search for Dynamic Targets with Uncertain Motion Models

This paper investigates the problem of search for moving targets when the target motion is poorly known. The approach taken is probabilistic, modeling the target motion with a discrete-state, discrete-time Markov chain-like model. The target motion across the discretized environment is described by a probabilistic state transition matrix. Because the target motion at each time step will not be known with certainty, this transition matrix may be poorly known, which could result in an incorrect allocation of the search resources. This paper presents a new algorithm that accounts for the uncertainty in the transition matrix. The algorithm uses an approach similar to particle filtering to stochastically simulate the uncertain state transition matrix, but approximates the posterior distribution with an analytical, closed-form distribution. By approximating with this closed form, resampling techniques normally used in stochastic sampling techniques are avoided, thereby making the proposed approach computationally very efficient. The new technique is evaluated in numerical simulations and the results show that considerable improvements in the search performance are possible by accounting for this uncertainty in the planning