A Gestural Language For A Humanoid Robot PDF Print E-mail
Written by Rizki Noor Hidayat Wijayaź   

Aaron Ladd Edsing er Submitted to the Department of Electrical Engineering and Computer Science on May 11, 2001, in partial fulfillment of the requirements for the degree of Masters of Science in Computer Science and Electrical Engineering

Abstract This thesis describes work done at the MIT Artificial Intelligence Laboratory on the humanoid robot platform, Cog. Humanoid research has long been concerned with the quality of the robot*s movement. However, obtainingthe elusive tempo and grace of the human motor system has proven to be a very difficult problem. The complexity of controlling high degree of freedom (DOF) humanoid robots, combined with insights provide by neurophysiological findings, has lead researchers to look at motor primitives (Williamson 1996) as an organizing methodology. We propose a data-driven approach to motor primitives in building a motor language for Cog. The proposed model is implemented on Cogand applied to the task of human motor mimicry.

Overview

This thesis describes work done at the MIT Artificial Intelligence Laboratory on the humanoid robot platform, Cog. Humanoid research has long been concerned with the quality of the robot*s movement. However, obtainingthe elusive tempo and grace of the human motor system has proven to be a very difficult problem. The complexity of controlling high degree of freedom (DOF) humanoid robots, combined with insights provide by neurophysiological findings, has lead researchers to look at motor primitives (Williamson 1996) as an organizing methodology. We propose a data-driven approach to motor primitives in building a motor language for Cog. The proposed model is implemented on Cogand applied to the task of human motor mimicry.

Approaches to Humanoid Motor Control

In this chapter we describe the issues encountered when developingmo tor control systems for robots and for humanoid robots in particular. The set of problems that are posed by humanoid robotics are in some sense unique to the field in general.

These problems both constrain and complicate research in this area. The scope of investigation for this thesis is motivated this set of problems. Humanoid robotics is an endeavor inherently concerned with realism. As researchers in the field, we are attemptingto model, simulate, and approximate what may myopically be considered one of nature*s greatest accomplishments. In building humanoid robots we are tacitly asserting the following: human level intelligence and interaction with the world requires both a physical presence in the world and a human morphology. As (Brooks, Brezeal et al. 1998) have noted, we have human intelligence by virtue of havinga human body.

However, it is not clear to what degree we need to model and mimic natural systems. The level at which we choose to approximate nature certainly varies depending on the area of investigation and on the purpose of that investigation. In the domain of humanoid motor control, this raises a number of issues. Clearly, we would like our robots to move in a human-like manner. Nature*s animals, regardless of our perception of native intelligence, never fail to astound us with the grace and dexterity of their movements. This overarching motivation has pushed researchers to a detailed study of biological motor control, ranging from physiological examination, to biomechanical models, to cognitive theories. Implementinga motor control system with a similar quality of movement on a humanoid robot can be considered one of the grand challenges of the field. It is still an open question, however, what method or approach is suitable for this goal. Researchers workingto wards this end have recognized that a simple recreation of human movement, as is done in the field of animatronics, is not enough in itself.

The aesthetic of robot movement is closely tied to the morphology of the robot, the electro-mechanical dynamics of the robot, and most importantly, a tight coupling between the robot*s environment, perceptual system, and motor system. Naturalism for humanoid robot movement must not just come from the well-coordinated execution of movement trajectories, but also from movement that is in appropriate response to the environment. Unfortunately, the best of our attempts have been hampered by the physical technologies available to us. Actuators such as DC servo motors are bulky, heavy, and consume inordinate amounts of power. The sensorimotor feedback available on our current systems pales in comparison to the massively parallel feedback employed in nature.

Regardless of our current technical limitations, there is still an interesting set of questions to explore. In fact, much of the grace and dexterity we admire in animals may not be a direct result of sophisticated musculature and sensory systems.

[download: 5,86 MB]