Low-Cost Localization for Educational Robotic Platforms via an External Fixed-Position Camera Drew Housten Dr. William Regli NSF Grant OCI NSF Grant SCI
Pre-College Educational Robotics Robotics is an excellent tool to teach AI, Engineering, Math, and Science Currently, educational system sophistication heavily depends on hardware cost LEGO NXT (Fairly Cheap, Limited) AIBO (Expensive, More Sophisticated) But, cost of the solution matters in pre-college education! Research does not follow the same trends Example: DARPA Urban Challenge was mostly a software problem
Pre-College Educational Robotics Hardware complexity of educational robotics is currently sufficient However, Software and System complexity of educational robotics is limited This problem can be addressed by building software tools to bridge the gap Software tools can be free to educators
Why Localization? Chose Localization as a starting point Currently many AI educational projects are limited because the robot does not know where it is Maze Following Navigation Searching Etc.
Problem of Localization Current solutions in research: Odometry Global Positioning Systems (GPS) LIDAR Sonar or Infrared Arrays Contact Sensors Fuducials or Landmarks Cameras Etc. Current solutions do not work well for education Expensive Complicated to use Does not work well in typical educational environments
CamLoc (Camera Localization) Goals of CamLoc Inexpensive solution to localization Simple to use Requires no hardware modifications Simplistic solution to support teaching the principles to students Decimeter-level accuracy in localization in an indoor environment
Necessary Hardware iRobot Roomba ($200) SparkFun Electronics RooTooth ($100) Computer ($500 - $2500) Webcam ($50-$150) Total Cost w/o Computer: ~$400
Technical Approach: Fusion of Odometry & Visual Tracking Topological Mapping: 1) Record Robot’s start position in the image frame 2) Make an action (point turn, drive) 3) Record odometry distance and heading traveled 4) Record Robot’s end position in image frame 5) Add an edge to the Topological Map Vertices are the image frame positions Localization: 1) Search through the Topological map to find a path between the initial position and the current position 2) Calculate the current position by simulating the actions to travel that path
Results from 3 Runs Square Circuit 39 Actions Meters Cloverleaf Circuit 50 Actions Meters Pseudo-Random 84 Actions Meters Mean Positional Error
Future Work Enhancements and Improvements to Approach Build a more complete toolkit to assist robotic educators Use the solution in a classroom setting Make the toolkit available for download at LearningRoomba
Questions?
Backup
Odometry vs. Topological Map
Vision Tracking Interface
Trial 1: Square Circuit
Trial 2: Cloverleaf Circuit
Trial 3: Pseudo-Random Path
Approach Goals Localization to decimeter-level accuracy Low-cost Solution Easy to configure / setup / use Elements of Solution Odometry Topological Map Image Tracking