Robust, real-time vision-based navigation and localization
Goal: Enable robots to autonomously navigate throughout an office or home without ever getting lost.
Together with Niklas Karlsson and Paolo Pirjanian of Evolution Robotics
and Dave Benson and Enrico Di Bernardo of idealab!, I have developed a vision-based navigation system
that enables a robot to localize itself with 10cm std. dev. accuracy.
Since the system is currently being patented, I cannot say much. But I can say this:
- - the only sensors are the robot's wheel odometry and a webcam
- - an accurate map of the environment can be generated after 15 minutes of driving
- - the system is very robust (has been extensively tested and never failed)
- - works reliably without ANY alteration of the environment
- - Patent being filed
- - yellow dotted curve : actual path
- - green curve : odometry alone
- - red curve : vision-based localization