An accurate and computationally efficient method for whole-body human modeling with applications in HRI

Abstract

Interactive robots are required to have minimal footprint on the shop floor and to be able to work in constrained areas while ensuring the safety of the humans. To this end, modeling of an unstructured environment including the humans is an indispensable part of the online control schemes deployed in such robots. In this regard, a new approach is proposed for generating an efficient model of the human body. This model takes advantage of superquadric functions to represent the human body more realistically than using primitive shapes, while offering minimal computational complexity and simplicity of further computations in comparison to the existing articulated models. This approach is also capable of incorporating various body postures and arbitrary arm configurations in the model. In addition, a new and intelligent sensory system, called floor mat, is introduced which can significantly contribute to generation of the proposed model, in a timely manner. The integration of the floor mat in a multi-sensory system for obtaining all required data for rendering the real-time 3D human model is then discussed. The use of superquadric-based human model in conjunction with the proposed sensing techniques provides an accurate, yet computationally efficient solution for safe human-robot interactions (HRI).