Cooperative anchoring : sharing information about objects in multi-robot systems

Sammanfattning: In order to perform most tasks, robots must perceive or interact with physicalobjects in their environment; often, they must also communicate and reasonabout objects and their properties. Information about objects is typically produced,represented and used in different ways in various robotic sub-systems. Inparticular, high-level sub-systems often reason with object names and descriptions,while low-level sub-systems often use representations based on sensordata. In multi-robot systems, object representations are also distributed acrossrobots. Matters are further complicated by the fact that the sets of objects consideredby each robot and each sub-system often differ.Anchoring is the process of creating and maintaining associations betweendescriptions and perceptual information corresponding to the same physicalobjects. To illustrate, imagine you are asked to fetch “the large blue book fromthe bookshelf”. To accomplish this task, you must somehow associate the descriptionof the book you have in your mind with the visual representation ofthe appropriate book. Cooperative anchoring deals with associations betweendescriptions and perceptual information which are distributed across multipleagents. Unlike humans, robots can exchange both descriptions and perceptualinformation; in a sense, they are able to “see the world through each other’seyes”. Again, imagine you are asked to fetch a particular book, this time fromthe library. But now, in addition to your own visual representations, you alsohave access to information about books observed by others. This can allow youto find the correct book without searching through the entire library yourself.This thesis proposes an anchoring framework for both single-robot andcooperative anchoring that addresses a number of limitations in existing approaches.The framework represents information using conceptual spaces, allowingvarious types of object descriptions to be associated with uncertainand heterogeneous perceptual information. An implementation is describedwhich uses fuzzy logic to represent, compare and combine information. Theimplementation also includes a cooperative object localisation method whichtakes uncertainty in both observations and self-localisation into account. Experimentsusing simulated and real robots are used to validate the proposedframework and the cooperative object localisation method.