Conveying information through non-verbal communication in robotics

Conveying information through non-verbal communication in robotics

Hameeteman, D.M.J.

Lopes, G.A.D. (mentor)

Mechanical, Maritime and Materials Engineering

Delft Center for Systems and Control

2016-04-20

Robots are becoming more and more involved in our daily life. In order to increase the acceptance of modern robotics in society even further, quite some improvements in human-robot interaction have to be made. Non-verbal communication is one such way of conveying information between robots and humans. In spite of its great potential, the current state of research is fairly limited. The goal of this research is to create a control framework that could generate movements with the flexibility to convey information by adapting the movement trajectories. Other studies have mainly researched non-verbal interaction using an optimized feedforward controller. Humans are however able to move without learning all movement sequences in advance, which is the motivation to solely focus on feedback controlled motions in current research. Considering the motion metrics traveled path, electrical power, gravitational torque and end-effector jerk, the proportional-derivative controller in the workspace proves to be best capable of producing the most human-like motions. The control framework for communicating non-verbally consists of an Operational Space Control hierarchy, combining the aforementioned controller and a novel method minimizing the gravitational torque during movement. Furthermore, human-likeness is even further increased by imposing joint limits based on the human extremes. These limits cause potential local minima, which prevent the controller from reaching the desired final position. Introducing an extra, unusual coordinate helps overcoming these minima taking into consideration the angle between current and final position. With these components included, the framework is able to reach any desired location taking into consideration human-likeness and maximum joint angle constraints. The Labanotation is used to describe the different motions to be conducted, the same way as is done with human dancing motions. Combining this representation with the Laban Time, Space and Weight Effort, it is now possible to conduct different movements. Professional dancers were asked to participate in an experiment, resulting in the relation between the end-effector trajectories and the Laban Basic Efforts: dab, flick, float and glide. These trajectories were used to optimize the gains within the control framework. The procedure was run for different initial and final desired positions, resulting in four sets of control gains corresponding to each of the four Efforts. Different motions are obtained using the four different control sets. Simulation shows that direct and flexible motions (Space Effort) are distinguishable by the shortest trajectories opposed to longer traveling distances respectively. High gravitational torque gains and low spring and damping gains result in more flexible motions. Furthermore, sustained and sudden motions (Time Effort) differ by the time the end-effector needs to reach the desired position. Sudden motions show to have higher spring-damping ratios with respect to sustained motions. In summary, the general control framework proposed by this thesis has the flexibility to generate motions either using the dabbing, flicking, floating or gliding Effort and therefore contributes to the acceptance of robots in society.

convey information
non-verbal communcation
Laban Efforts
Operational Space Control
human-likeness
gravitational torque
robotics

http://resolver.tudelft.nl/uuid:b6e101a0-9c61-4e75-8a64-dae3759f4216

Student theses

master thesis

(c) 2016 Hameeteman, D.M.J.