Synthetic Robotic Language Acquisition by Observation

Abstract

Herein we describe work that addresses issues of inter-agent communication skill acquisition by observation in a society of interacting agents. A bee-like society of agents is introduced (most adjacent to these experiments in ethological terms) and an architecture for enabling agents to acquire new skills (learn) by observation is proposed. The architecture is based on a combination of connectionist models, namely Kohonen self-organizing networks and back-propagation networks. A synthetic, movements-based robotic language is devised as a means of verifying the architecture. The architecture enables the agents to learn both how to understand and how to use, by generating new messages, a simple language that can be seen as the agents' symbol grounding of semantic knowledge. The tools used for implementing and testing this work included multiagent simulators and autonomous robotic vehicles as well as other specialized hardware like cameras and transputer boards for vision-based robot posi...