One year ago today, I wrote a post about research at Carnegie Mellon that purported to allow a robot to sort items. As I noted, this is a challenging task.
But Carnegie Mellon isn't the only entity that is working on this. There's also Willow Garage:
Robotics company Willow Garage has started a two-year project to work with institutions from around the world on new applications for its robot: the PR2. Each of 11 teams will work on their own projects, but will share their code with each other and the rest of the world. Everything created will be open-source....
One of the teams is from MIT.
Folks at MIT's CSAIL lab, meanwhile, will work on object recognition and putting away groceries.
I searched the MIT CSAIL website and could find no mention of groceries, but perhaps the project described here may be related to the Willow Garage effort.
Feature-based object recognition for objects that do not have features.
Tomatoes are currently recognized using a feature based approach. In a learning step a set of features are trained that are reliable indicators of the object that needs to be found in the image. In a convolution step unknown images are convoluted with each individual feature, which then “vote” for the object location in the image. It turns out that this approach works well for objects that have distinguished features or textures but less so if distinctive features are rather part of the background than the actual image – as it is the case for tomatoes. We would thus like to extend the current algorithm to better cope with this particular type of objects.
This project requires solid programming skills in Matlab and C and a good understanding of mathematical concepts such as convolution and probability. Experience with OpenCV is advantageous, but can be acquired during the project.
Contact:
Daniela Rus(rus@csail.mit.edu) and Nikolaus Correll (nikolaus@csail.mit.edu)
I hope that the designers are taking care to ensure that the robots do not squeeze the tomatoes.
