With the services that autonomous robots are to provide becoming more demanding, the states that the robots have to estimate become more complex. In this thesis, a vision-based, probabilistic state estimation method for large and complex states is developed and applied to autonomous mobile robot applications. The proposed method extends the state-of-the-art in state estimation in two important ways. First, it demonstrates how the estimation of states in complex and ill structured state spaces, spanned by more than 60 parameters, can be achieved. The state estimation problem is made feasible by decomposing it into several loosely coupled subproblems. Each subproblem is solved by a task specific state estimator. In particular, this method enables the mobile robots of a team to estimate their own positions in a known environment and to track the positions of independently moving objects at frame rate. The second contribution is the investigation of the use of cooperation, i.e. the exchange of observations and state estimates between robots, for improved state estimation. The state estimators of the robots are extended to use the information provided by other robots as evidence. This information is shown to increase the accuracy, reliability and completeness of the state estimation process. In particular, it is demonstrated that cooperation enables robots to determine their poses and the positions of further dynamic objects more accurately, to track temporarily occluded objects successfully, and to obtain a complete view of the surrounding environment. The method is empirically validated based on experiments with a team of autonomous robots equipped with off-the-shelf computing hardware and sensory equipment within the RoboCup scenario. It was applied during four RoboCup world championships. The collected experimental data, from two competitions covering more than four hours net operation time, is analysed.
«
With the services that autonomous robots are to provide becoming more demanding, the states that the robots have to estimate become more complex. In this thesis, a vision-based, probabilistic state estimation method for large and complex states is developed and applied to autonomous mobile robot applications. The proposed method extends the state-of-the-art in state estimation in two important ways. First, it demonstrates how the estimation of states in complex and ill structured state spaces, s...
»