Scavenging with a Laptop Robot

We designed and implemented a robot to participate in the AAAI '05 Scavenger Hunt Competition. The Scavenger Hunt is a showcase for robotics researchers to demonstrate the abilities of their robots to find and identify objects in a dynamic environment.

We demonstrated that a low-cost sensor-limited robot could be competitive against well-equipped conventional robots. Focusing on computer vision and software processing allowed us to easily detect and recognize objects. Furthermore, by combining visual data with a standard laser pointer, we were able to construct a rangefinder for under $50.

During the competition, we demonstrated the ability to identify a beachball and deliver it to a specified location. We differentiated between an orange cone, an orange stuffed dinosaur, and an orange bowl. We also successfully followed a series of arrows. Additionally, our system had the ability to recognize blue and green bowls as well as racquetballs in a variety of colors.

Our performance at the conference earned First Place in the Scavenger Hunt and a Technical Innovation Award for "Overall Excellence in a Fully Autonomous System."

Pictures and video of our robot are available at http://ben.st.hmc.edu/summer05/images/

• AAAI '05 Entry Submission
• AAAI '05 Conference Poster (16Mb)
• AAAI '05 Technical Paper
• SCCUR '05 Poster Submission (20Mb)
• IASTED Robotics and Applications '05 Technical Paper

Here is a zip file with everything: AAAIsourcecode.zip

If you have questions about our implementation, algorithms, or architecture, please contact us.

• er1.py
  This implements a driver for Evolution Robotics' ER1 platform. It can be imported as a standard Python module or run as a stand-alone program to provide a client-server interface to the ER1. er1.py is written in Python and is provided as a flat Python source file.
  Questions should be directed to Ben Tribelhorn or Alan Davidson.
• MapTool.zip
  MapTool provides localization and path-planning algorithms for a given map. In addition, it provides the ability to place arbitrary objects on the map. MapTool is written in C++, and is provided as a Microsoft Visual Studio .NET project.
  Questions should be directed to Mac Mason or Susanna Ricco.
• VideoTool.zip
  VideoTool does connected-component and shape analysis after segmenting based on color. These color definitions can be changed in realtime and can be loaded from and saved to a file. VideoTool uses Intel's OpenCV library to capture video from a camera. Finally, VideoTool provides the processing for our custom laser rangefinder. VideoTool is written in C++ and provided as a Microsoft Visual Studio .NET project.
  Questions should be directed to Alan Davidson or Ben Tribelhorn.
• oo_remote.py
  This provides the controlling architecture for our robot. It communicates with MapTool and VideoTool via a client-server model, and imports er1.py as a Python module. oo_remote.py implements a class-based finite state machine. oo_remote.py is written in Python and is provided as a flat Python source file.
  Questions should be directed to Susanna Ricco, Ben Tribelhorn, or Alan Davidson.

Harvey Mudd College is an undergraduate institution whose aim is to educate scientists, engineers, and mathematicians with a firm understanding of the impact of their work on society. It is located in Claremont, California and is a member of the Claremont Colleges.

Our team is composed of Alan Davidson (Joint CS-Math '06), Mac Mason (CS '06), Susanna Ricco (Joint CS-Math '06), and Ben Tribelhorn (CS '07). All four of us plan on attending graduate school upon graduation from Harvey Mudd. Our advisor is Prof. Zach Dodds. Comments or questions about our project or code can be directed to any or all of us.

We would like to thank the NSF for the grant that funded this research as well as the HMC CS Department and AAAI '05 for providing additional funding. We would also like to thank Yau Lam Yiu and Jason Campbell for access to their source code. Prof. Erik Spjut of the HMC Engineering Department lent his technical expertise to the design of our sonar circuit.