Can robots recognize common Marine gestures?

This paper provides a different method for human-robot interaction and further encourages communication between human users and their robotic counterparts. The focus of our work is to develop a human-robot communication system that is not easily detectable and increases stealth when necessary. The human-robot interaction system we propose involves hand signals. Hand gestures are a common modality of communication humans use with each other. Likewise, hand commands are used by the Marine Corps to convey information to each other without speaking. We analyze common Marine gestures so that similar commands can be used to direct a robot out in the field. In this paper, we have selected important hand or body gestures used by the Marine Corps. We then identify distinguishable features for the different gestures. This includes position of joint variables as well as velocity and acceleration terms. Once ideal models of the gestures are designed, experimental data is gathered. Presently, we are comparing two different machine learning methods that can be used to identify a specific gesture. The two methods we are comparing are Bayesian networks and neural networks. This paper provides the background and structure of our experiments. Then, both models are discussed and experimental results are included. Finally, we make a comparison of the effectiveness of the two methods of interest.