Temporal arm tracking and probabilistic pointed object selection for robot to robot interaction using deictic gestures

Abstract

This paper discusses advancements on alternative communication methods between robots utilizing deictic gestures and body language. The provided approach boosts the performance of the formerly developed ROS-oriented system and grants increased accuracy and robustness. More precisely, the paper presents: a) a temporal tracking technique responsible to detect and track the arm of the pointing agent and b) a probabilistic pointed object selection method. On one hand, a model registration approach based on the Interactive Closest Point (ICP) algorithm, continuously tracks a detected pointing arm over time. On the other hand, a probabilistic method weights a number of features to estimate which of the available objects is the pointed one. The provided results indicate that the enhanced system outperforms the formerly developed approach and efficiently tackles any identified problems, such as the correct pointed object selection in ambiguous situations.