Improved Vision-based Robot Navigation Using a SDM and Sliding Window Search
Mateus Mendes, A. Paulo Coimbra, Manuel M. Crisostomo · 2011
Robust and fast vision-based robot navigation is a long sought goal, which requires comparing the robot's current view with a database of visual memories. The technique described in the present paper uses a Sparse Distributed Memory (SDM) to store paths described by sequences of images, and a sliding window to narrow the search space for real-time operation. The use of the sliding window greatly reduces the processing time and the number of prediction errors. The use of a short-term memory confers on the robot the ability to still solve the kidnapped robot problem. The Sparse Distributed Memory is a kind of associative memory suitable to work with high-dimensional binary vec- tors, thus being appropriate to store long sequences of images. images it has to store grows continually. The memory requirements and the processing time increase in propor- tion to the database size. It should be noted, however, that modern evidence indicates that the human brain functions in a similar manner: it is a huge amount of memory, used to store sequences of events that will lead future analysis and actions (2, 3). The images alone are a means for instantaneous locali- sation. View-based navigation is almost always based on the same idea: during a learning stage the robot learns a sequence of views and motor commands that, if followed with minimum drift, will lead it to a target location. The robot is later able to follow the learnt path by following the sequence of commands, possibly correcting the small drifts that may occur.