Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 149729

149729 - Development and Evaluation of a Teaching-Learning-Prediction-Collaboration Framework for Human-Robot Collaborative Advanced Manufacturing 

The rise in the employment of robotic systems across modern industries – particularly in advanced manufacturing contexts – has prompted the research and development of more advanced and capable systems. Among these systems are those that seek to blur the line between human-only and robot-specific tasks, merging them through the implementation of collaborative robotics that permit the worker and machine to complete the task at hand with each other’s help. By making use of each other’s strengths, tasks may be completed quicker, safer, and easier when compared to traditional methods of setting them to do their own dedicated jobs. Furthermore, collaborative systems have recently been receiving attention in the field of robot learning by demonstration. In this method of collaboration, robots observe their human counterpart perform a task and learn how to perform it. In this way, they may learn their teammate’s working preferences and how precisely to perform their job when operating with a specific user. What remains to be seen, however, is just how these human workers feel about interacting with a system that attempts to dynamically learn from their demonstrations, as well as whether it is preferable to traditional systems. Motivated by these issues to determine whether these learning and teachable collaborative systems are perceived as superior to traditional ones and how they may be further improved. This research details our study into a collaborative system that makes use of this Learning from Demonstration methodology to develop a synergistic Teaching-Learning-Prediction-Collaboration (TLPC) framework. Additionally, to evaluate the performance of the proposed framework, 109 user studies are conducted with/without TLPC in human-robot collaborative manufacturing contexts. The experiment was performed by designing a collaborative setup in which users would interact with a Franka Emika Panda robot to complete an assembly task, which they first demonstrated to the robot. The robot’s vision system captured this demonstration data and modified its behavior accordingly by constructing task knowledge that may be represented by a finite state machine. Experiment participants expressed two separate methods of assembly for the robot to learn, and following the “teaching-learning” phase, the robot assisted with the task in its learned method. Following this, participants were instructed to attempt the same operations using the traditional non-learning approach. A survey was provided after each interaction with seven evaluation metrics (task efficiency, collaboration safety, coding efforts reduction, collaboration fluency, collaboration sociability, robot response speed, and overall comfort) ranked with a Likert scale to gauge their perceptions of each approach. From the survey data collected from the experiment’s participants, it suggests a strong preference for the TLPC framework. That is to say, users felt a much stronger appreciation for the ability to demonstrate and instruct the robot on how to perform their decided assembly task than they did for the typical hard-coded method. This conclusion was established through the use of 3 separate t-tests (Student’s, Welch’s, Paired) performed on each evaluation metric between the two methods. These tests all returned p-values far below the designated significance level of α=0.05, providing more than enough evidence to reject the null hypothesis that the differences between the two approaches were not statistically significant. Though ratings were high for two of the evaluation metrics of “task efficiency” and “collaboration fluency,” it is the goal of this research to determine solutions to improve on these. Thus, future research will focus on designing and implementing these improvements.

Presenting Author: Garrett Modery Montclair State University

Presenting Author Biography: I am a senior undergraduate computer science major studying at Montclair State University. I currently have the pleasure of working under the supervision of Dr. Weitian Wang in the MSU Collaborative Robotics and Smart Systems laboratory, where my work is primarily focused on the refinement and research into the human-robot working dynamic. Working with these systems has opened my eyes to the vast field of robotics, machine learning, and human-robot interaction, and I am immensely grateful to have the opportunity to conduct the research that I do.

Authors:

Garrett Modery Montclair State University
Weitian Wang Montclair State University