Transferring embodied concepts between perceptually heterogeneous robots

Abstract

Abstract-This paper explores methods and representations that allow two perceptually heterogeneous robots, each of which represents concepts via grounded properties, to transfer knowledge despite their differences. This is an important issue, as it will be increasingly important for robots to communicate and effectively share knowledge to speed up learning as they become more ubiquitous. We use Gärdenfors' conceptual spaces to represent objects as a fuzzy combination of properties such as color and texture, where properties themselves are represented as Gaussian Mixture Models in a metric space. We then use confusion matrices that are built using instances from each robot, obtained in a shared context, in order to learn mappings between the properties of each robot. These mappings are then used to transfer a concept from one robot to another, where the receiving robot was not previously trained on instances of the objects. We show in a 3D simulation environment that these models can be successfully learned and concepts can be transferred between a ground robot and an aerial quadrotor robot.

Architectural design and support for knowledge sharing across heterogeneous MAST systems

“…This paper presents the transition from the theoretical research developed earlier (e.g., [6][7][8][9][10]) to a practical working architectural implementation of Gardenfors conceptual spaces to a team of completely perceptually heterogeneous robots. Using sensors being developed by the Army's Microautonomous System Technology program and their surrogates, the abstract sensor layer (ASL) and conceptual space component of the overall system architecture has been implemented and tested, demonstrating the viability of this concept in the context of a biohazard search and detection mission.…”

“…We continue to exploit our novel approach to this problem that leverages our significant prior work [6][7][8][9][10] in the conceptual spaces framework developed by Gardenfors [4,1]. In this paper we focus on the architectural design and implementation issues associated with widely heterogeneous micro-robots that are tasked for specific missions.…”

Using conceptual spaces to fuse knowledge from heterogeneous robot platforms

“…This paper builds upon our previous work and shows that a hierarchical concept representation can be used to facilitate the fusion of concepts once the robots determine which of the representation's lower-level building blocks are shared. Previously, we have shown that the mappings between properties of two robots can be learned using real-robot data obtained in a shared context [5], and we have demonstrated the transfer of concepts from one robot to another in realistic 3D simulation experiments [6]. We have also shown that the abstraction of data through higher-level properties aids both learning and knowledge transfer, in real experiments where support vector machine classifiers are used to combine properties and classify objects [7].…”

“…We begin by formally describing the framework and demonstrating how raw sensory data is abstracted into higher levels properties, which are then combined to represent concepts [5,6]. Validation of this framework for the classification of real objects using sensor obtain from real-robot experiments are described in the next section.…”

Cooperative Knowledge Based Perceptual Anchoring

“…Despite the clear focus on the semantic aspect, the works mentioned above are lacking a cooperative multi robot aspect, which we try to also capture in our current work. This specific aspect of transferring embodied concepts between perceptually heterogeneous robots has been studied in 19 , where the different perceptually heterogeneous embodiments communicate and effectively share knowledge that they have learned through individual experiences. Their approach is based on communication and knowledge exchange via social interaction, where properties of the world can be grounded differently in different robots.…”