Extraction of salient features for mobile robot navigation via teleoperation

This paper presents a method to extract salient features from sensory-motor sequences for mobile robot navigation via teleoperation. Salient feature extraction consists of three steps: teleoperation, offline association, and evaluation. First, the mobile robot is teleoperated in an environment along a path several times. All sensory data and motor drive commands are recorded. During an offline association step, these sensory-motor sequences are partitioned into episodes according to changes in motor commands. Salient features are then extracted by using two statistical criteria: consistency and correlation with the motor commands within the episode boundaries. Finally, these features are used to drive the robot in the learned environment. Some experiment results are also presented.