Deon Sabatta
|
Better navigation using less data for autonomous robots
CSIR robotics researcher Deon Sabatta has developed a method to reduce the amount of data storage required for autonomous robots during navigation using persistent Scale Invariant Feature Transform (SIFT) features.
His approach is based on topological mapping. Sabatta says, "The concept of topological mapping was introduced into the field of robotics following studies into human cognitive mapping. Topological maps store only unique or distinct areas and the relationships between them. This is similar to the way humans store information of their environments."
Sabatta says that most problems in autonomous robotics require systems that develop an environmental map and then localise themselves within this map. Topological mapping allows for the use of image properties that are not specific to an exact location and also permits some variability in the environment. SIFT features provide descriptors that enable topological mapping based on image similarity. "Our method is similar in that topological regions are defined as a collection of SIFT features. However, we have achieved a reduction in the number of stored features by only considering persistent features that exist in the vicinity of a topological region."
While other methods space regions equally in time or distance, this new technique also allocates topological regions conservatively, based on the number of visible features. This region definition coupled with the selection of only persistent features from the myriad of features usually identified within an image, limits the total amount of information required to map an environment by 70% relative to other methods storing SIFT descriptors indiscriminately. This reduction in the number of features stored also results in an improvement in performance over related algorithms.
Explaining the methodology, Sabatta says, "We collected local image features extracted from a panoramic image (using omnidirectional camera images) of the environment. We then grouped these features based on their persistence within the local environment." He explains, "In the construction of the topological map, regions usually share some features of nearby regions. Navigation through this map can then be accomplished by following a trail of common features between the regions en route to the final destination. Through careful selection of features, the amount of data required to define a region is reduced and rapid localisation is possible."
Sabatta adds, "This development is set to make a significant impact in the area of studies into autonomous robots, specifically navigation. My research endeavours will continue to aim at methods of enhancing this pivotal area in robotics research."
Enquiries: CSIR Communication
|
