Self Body Mapping in Mobile Robots Using Vision and Forward Models

The work presented in this paper aims at providing an agent with basic capabilities leading towards navigation through self body-mapping using a vision system. In particular, we are interested in the study of forward models which code for the sensory consequences of an agent's self-produced actions. The research is developed in the framework of cognitive robotics and embodied cognition. The agent is a robot we let interact with its environment to know the free space around it from re-enaction of sensory-motor cycles predicting collisions from visual data. From the disparity map the robot associates intensity regions with motor commands in order to predict distances to objects reported by a tactile sensor in self-motion coordinates. In order to form this multimodal association we use a forward model, coded as a system of neural networks and trained with data coming from random trajectories executed by a Pioneer 3-XD. The resulting forward model allows the agent to navigate avoiding undesired situations by performing long term predictions of the sensory consequences of its actions. The experiments validate the hypothesis that this model allows for a basic self body-mapped navigation capability.