The Undergraduate Robot Research Team

Winners of the Robot Host Event at the AAAI-2002 Mobile Robotics Competition!

Road Rules, CS Style! Edmonton Bound - to AAAI!. From left to right: (front) Eric Meisner, Jenine Turner, Tori Sweetser, Jon Schmid, and Ben Atwood. Back: Tom Kollar. Missing from the picture: Dave Feil-Seifer (stuck in an elevator in the dorm!) and Mike Isman.
Thanks to the ambition and enthusiasm of undergraduate Jonathan Schmid '03, 9 undergraduates participated in the AAAI (American Association for Artificial Intelligence) 2002 conference July 28-August 1, 2002 in Edmonton, Alberta. The 4600 mile road trip was the culmination of over a year of work. Jonathan's desire to compete in the robot contest got the ball rolling. He attended AAAI last summer to see what the competition was like, ready to plan a strategy for this year's conference. With the aid of faculty advisor Chris Brown, Jon & his hand-picked team began to implement the plan. Eleven students did independent research on the robot (Mabel, the Mobile Table) for the fall semester, and nine were on board for the spring semester. The faces & names morphed again to the summer team of competitors (David Feil-Seifer, Jonathan Schmid, Ben Atwood, Michael Isman, Thomas Kollar, Eric Meisner, Tori Sweetser, Jenine Turner).

Working to create a human interaction agent on a mobile robotic platform, the team divided the task of making an optimal robot into Conversational Speech Interaction, Active Vision, and Autonomous Navigational Control.

Conversational Speech Interaction involves phoneme recognition in a noisy environment, and parsing natural language into a concrete set of percepts. This information allows for an appropriate multi-modal response using a combination of speech, food manipulation, and navigation behaviors. We accomplish this using the Sphinx Speech Recognition System developed by CMU augmented by a digital filter. We employ a directed speech recognition microphone, which is actively pointed towards the speaker, using face tracking and a pan-tilt-zoom camera. To accomplish language understanding, a specially designed grammar-based parsing technique is under development.

The vision component's purpose is to provide the navigation component with real-time visual percepts including:

  • estimated number of people in the closest patron group
  • best approach angle in field of view for closest patron group
  • patron face-region detection and tracking
  • nametag-region detection
  • scene character segmentation and classification (ANN)
  • patron face recognition (HMM)

We have demonstrated robust and successful implementations for many of the above systems. We have done preliminary work on the features marked "under-development", and expect to include them as part of our final entry.

Autonomous Navigation Control involves creating a robust model for navigating around a crowded room while retaining the ability to return to a base station. We use sonar-based obstacle avoidance for robust navigation. To achieve path planning and execution, we employ a trained waypoint system using wheel counters.

Additional students who participated in earlier work on the team include Aaron Zschau '02, Habib Chaudhury '03, Scott Cragg '04, (Vision Team), Ross Camara '03 (Control Team), and Diana Calarese '03 (Language Team).

Related Links

Mabel, The Mobile Table (July, 2002)


Maintained by Marty Guenther. Last change December 19, 2002.
Please report problems with this server to www@cs.rochester.edu