May 31, 2013 | Atlanta, GA
Ayanna Howard and two of her Ph.D. students, Hae Won Park and Richard Coogle, took the first place prize at the DARwIn-OP Humanoids Application Challenge at the 2013 IEEE International Conference on Robotics and Automation (ICRA 2013). The conference was held May 6-10 in Karlsruhe, Germany.
Dr. Howard and her students are affiliated with the School of Electrical and Computer Engineering (ECE) at Georgia Tech. Their team, dubbed as “Angry DARwIn: Framework for Human-robot Task Collaboration on a Shared Tablet Workspace,” demonstrated their robot learning from a human demonstrator how to play a game of Angry Birds. Based on Ms. Park’s Ph.D. thesis work, “Task Learning Policies for Collaborative Task Solving in Human-Robot Interaction,” the key elements to long-term engagement during human-robot interaction are the robot’s capability to learn and engage in new tasks through human demonstration and a defined workspace that provides content that both the human and robot participants can use and enjoy.
The framework that the team proposed for the competition provides a case-based reasoning approach to robot learning, a human reasoning methodology that retrieves the solutions of problems observed in the past, and reuses the solutions to respond to emerging similar problems. Researchers using the framework can model their own task as tablet apps and configure the framework settings to teach the task to their robots. The framework provides tools for modeling the task policy through training the parameters for case-retrieval function that maximizes the rewards. During the demonstration, the team’s robot, Angry DARwIn, was able to quickly learn the strategies to solve each level of Angry Birds after less than three or four demonstrations from the teacher. Angry DARwIn not only learns to play the game, but also picks up verbal and emotional behaviors that the human teacher demonstrates. The robot then converts the learned emotions into his own verbal and gestural primitives that merge with his task behavior. The team anticipates using Angry DARwIn as a mediator to research how social interaction emerges during human-robot interaction.
This award-winning work originated from the Human-Automation Systems Lab, led by Dr. Howard, who is the Motorola Foundation Professor in ECE. Their prize included $20,000 worth of the humanoid robot platform DARwIn-OP from ROBOTIS, robot simulator Webots Pro from Cyberbotics, and Labview Robotics Software from National Instruments.
In addition to the competition, Angry DARwIn was voted the best robot application through its promotional video and was awarded a $1,000 gift certificate from ROBOTIS. The video can be viewed at http://youtu.be/eCKInnYDB2E.