Back to Kyle Reed's page in LIMS.
Last updated by KBR on 5/10/07.
Throughout this thesis, I will discuss my experiments on physical interaction and negotiation between two agents working on a target acquisition task. First, I will evaluate human-human physical interaction. I will show that dyads are faster than individuals, despite applying larger forces. By analyzing the interaction forces through a jointly controlled object, I will reveal a distinctly different completion strategy for dyads that is not available for individuals. Second, I will look at human-robot interaction. By simulating human-human interaction, a robot can surreptitiously replace one of the human partners leading to improved disturbance rejection.
Original found in:
Kyle B. Reed. "Understanding the Haptic Interactions of Working Together," Ph.D. Thesis, Northwestern University, June 2007. [pdf]
Executive Summary:
A basic form of human interaction is the physical cooperation necessary to perform a manual task with others. Physical cooperation represents a communication channel distinct from facial expression, gesture, and spoken language, yet one that has been much less studied. Shared physical tasks require participants to adapt, anticipate, and react to each other's forces and motions and are often performed in the absence of any explicit verbal communication. An understanding of how two people physically cooperate, compromise, and guide one another is a fundamentally important aspect of human motor control and it may allow improved cooperation and physical communication between a person and a machine. The forces and motions involved in haptic communication may be coupled directly limb-to-limb, or via a mutually grasped object, which is the focus of this thesis.
As an initial foray into human-robot-human interaction, I devised the simplest experiments to reveal novel effects that arise in dyadic motion control. I designed and built a one-axis motion configuration in which two subjects participate symmetrically and communicate solely through physical interaction. The crank has three possible inputs to control the position: two handles that subjects manipulate and one motor that can apply a torque directly to the crank.
When a person completes a task individually, they necessarily apply all the required force. Adding a partner allows new possible completion strategies where each member can apply less than the total required force. A robotic partner can similarly aid in task completion. If the robotic partner is correctly designed to follow human-human communication standards, a robotic partner can surreptitiously replace one of the human partners.
My experiments reveal a dyadic completion strategy surprisingly different from that of an individual. When two people work together, each member actually applies significantly more force than an individual does. Some of this force is in opposition to their partner and some is in cooperation. The opposition force is internal to the dyad and results in no crank motion. I call this dyadic-contraction because it resembles co-contraction in an individual. Dyadic-contraction occurs when each subject applies an equal and opposite force to the their handle. I found that many dyads exerted a significant dyadic-contraction force, in some cases over 40 Newtons. I believe this force is predominantly for stabilizing the interaction of the dyad and also as a means for the members to communicate, since other channels are absent.
Communicating solely through the forces and motions of the crank, the members of a dyad were able to negotiate a distinctly different strategy than an individual. Both members used their past knowledge and collective force to cooperate throughout different task phases. I call this cooperation specialization, which consists of dyads temporally dividing the task such that one member takes control to accelerate while the other member takes control to decelerate the crank.
I implemented the human-human interactions discussed above into a robotic partner by simulating dyadic-contraction and specialization. Surprisingly, the robot was unable to elicit specialization in the subjects, but it was able to make subjects believe they were interacting with a person as opposed to a robot. I found that the human-robot pair was better at disturbance rejection than two humans working together.
Despite the opposition forces described above, the completion time of dyads was significantly faster than individuals. When a human worked with a robot simulating a specialized partner, the performance was similar to a single person completing the task. Subjects working with the robotic partner performed faster if they perceived the partner to be a human rather than a robot.
The culmination of my results shows that humans are able to cooperate on tasks that take a significant amount of time, such as specializing their forces based on task phase and improved task performance, but humans are less able to cooperate on rejecting quick force perturbations. A person working with a robotic partner programmed to simulate human-human interaction showed the opposite. Even though subjects believed the robot was a human partner, the robotic partner did not elicit the human-human completion strategy in a human. However, simulating dyadic-contraction did improve the disturbance rejection characteristics of a human-robot pair.