Autonomous Robot Relocalization with Motion Constraints

With the current interest in autonomous cars, the key ingredients of these unmanned vehicles are the mapping, localisation, and route planning modules. The mapping is closely coupled to localisation: to obtain an accurate map, localisation is required, and for precise localisation, a reference map is usually considered. In this work, we narrow the focus to the relocalization problem: re-estimating the pose of a vehicle in the presence of uncertainty in the perception model. For perception, we consider the input from a motion-constrained camera, typically mounted looking toward vehicles. For the pose estimation, we consider the motion constraints such as limited acceleration and turning in a Bayesian context.