I proposed and led this work through my time at the Robotics Perception, and Manipulation lab at the University of Minnesota, instructed by professor Karthik Desingh.

The work aimed to explore the use of exisitng table-top task learning methods for efficient mobile manipulation task learning. I progressed this work to the point of demonstration; I was able to use language commands to instruct the lab’s Boston Dynamics Spot robot to perform the base task of dragging a chair.

Much effort was put into achieving an acceptable success rate; the voxel space was registered to a static frame, external cameras were used to suppliment the view, feature extractors were used to add meaningful features to the voxel space, etc..

The conclusion of this work was that, even with much adaptation, voxel-based table-top models might not be ideal for efficient mobile manipulation task learning.