Robotics

Perception Challenge has next-generation robots in its sights

View 4 Images
"Robot's eye view" showing how some common household objects appear through the vision system being used in the Perception Challenge - although the objects are fuzzy because the cameras have limited resolution, the images do provide information on depth (Image: Willow Garage)
Recognition and 6 degree of freedom pose of an Odwalla juice bottle found in clutter
The PR2 robot fitted with a Kinect sensor
"Robot's eye view" showing how some common household objects appear through the vision system being used in the Perception Challenge - although the objects are fuzzy because the cameras have limited resolution, the images do provide information on depth (Image: Willow Garage)
The 35 known objects that must be identified
View gallery - 4 images

Despite all the breakthroughs in the world of robotics, we still seem to be some way off the kind of advanced robots that can autonomously carry out a variety of tasks in unstructured and cluttered environments. One of the key bottlenecks holding back the development of such next-generation robots is how robots perceive the data gathered from their various sensors. Willow Garage, the Californian robotics company behind the PR2 open platform robot, has teamed up with the National Institute of Standards and Technology (NIST) to launch an international "perception challenge" with the goal of encouraging improvements to sensing and perception technologies for next-generation robots.

Willow Garage and NIST see robust robotic perception as a core enabling technology for robots that can be used more complicated applications such as in-home assistance for the elderly, search-and-rescue operations and advanced manufacturing. For such applications robots need to be able to reliably identify objects and determine their position accurately.

To this end, the Perception Challenge will ask competitors to develop code to recognize the identity and pose of 50 common objects, of which 35 will be known beforehand. The code will need to run with the open source ROS (Robot Operating System) on a PR2 robot that has been fitted with a Kinect sensor. So the contestants don't have to learn ROS, they will be provided with a code stub that will provide the 2D image and 3D point cloud data captured with the Kinect sensors.

The PR2 robot fitted with a Kinect sensor

Entrants will compete in a set of increasingly difficult rounds, starting out with identifying single objects and moving up to recognizing multiple objects and their pose within a cluttered scene. The top performing code will then be demonstrated on a PR2 robot grasping the items. The prize money will grow exponentially with each round to reflect the increasing difficulty, topping out at US$7,000 to be awarded for the successful completion of all tasks within the allotted time. Second place takers will eligible for $3,000, and third place $2,000.

As well as spurring improvements in robotic perception, Willow Garage and NIST hopes the techniques and metrics demonstrated in the Perception Challenge will provide the foundation for new standards and test methods for measuring perception system performance. They also plan to increase the difficulty of the competition every year.

The inaugural Perception Challenge will take place at the IEEE International Conference on Robotics and Automation (ICRA) 2011, which will be held in Shanghai, China, on May 9-13, 2011. The deadline for entering the competition is April 15, 2001, with final submissions due by May 1, 2011.

View gallery - 4 images
  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
0 comments
There are no comments. Be the first!