WSEAS Transactions on Computer Research
Print ISSN: 1991-8755, E-ISSN: 2415-1521
Volume 4, 2016
A Novel Approach to Lifelong Learning for Robotic Response to Gesture
Authors: , , , , , ,
Abstract: For unskilled or impaired users, intuitive interfaces are seen as key to the wide adoption of assistive robotic agents. Interfaces based on gesture offer a natural communication modality for Human-Robot Interaction which facilitates ease of use for the target population. However, as the abilities of the user change with time, their capacity to properly perform a given command choreography may degrade. Therefore, it is critical to the longterm success of the interface that it learn the behavioral intentions of the user and adapt accordingly throughout the lifespan of the agent. This paper describes a clustering approach to an agent’s lifelong learning based on the Growing Neural Gas algorithm. A simulated user issues reinforcement feedback so as to progressively refine an agent’s actions toward a goal configuration. Several network distance metrics for neighborhood learning within connected clusters are compared for their relative burden to the user-trainer in terms of speed of convergence. It is shown that the proposed method learns new gestures while retaining past learning with few training iterations.
Search Articles
Pages: 138-152
WSEAS Transactions on Computer Research, ISSN / E-ISSN: 1991-8755 / 2415-1521, Volume 4, 2016, Art. #16