Timezone: »

High-Accuracy 3D Sensing for Mobile Manipulators
Stephen Gould · Morgan Quigley · Siddarth Batra · Ellen Klingbiel · Quoc V Le · Andrew Y Ng

Tue Dec 09 07:30 PM -- 12:00 AM (PST) @ None

Our demonstration shows the advantages of high-accuracy 3d scanning to the tasks of object detection and door opening. Specifically we show how to obtain depth data from a 3d line scanner directly in the image frame of reference. This allows us to associate a ""depth channel"" with each image pixel, i.e. each pixel is represented as a four dimensional vector comprising red, green, blue and depth. We then show how, using this representation, we can get near perfect door opening and object detection performance for a selection of common household/office objects. This level of accuracy on a mobile robot in uncontrolled environments have never been achieved before.

Author Information

Stephen Gould (ANU)
Morgan Quigley (Stanford University)
Siddarth Batra (Stanford University)
Ellen Klingbiel (Stanford University)
Quoc V Le (Stanford)
Andrew Y Ng (DeepLearning.AI)

More from the Same Authors