Robot Navigation Using View Sequences and Sliding Window Search

In previous work the authors developed algorithms to navigate a robot based on sequences of visual memories stored into a Sparse Distributed Memory—a kind of associative memory suitable to work with high-dimensional binary vectors. In that system, the robot’s localisation is based on similarity between the robot’s view and one previously stored image. In that system, prediction errors occur from time to time, when the robot’s captured image retrieves a wrong image from the memory. This paper describes a method in which the number of prediction errors is substantially reduced, through the use of a search sliding window.