AnyPick DraKe: A Behavior-Tree-Driven System for Open-Set Object Segmentation and Manipulation

AnyPick DraKe was the final project for MIT 6.4212 Robotic Manipulation.

Here is a working demonstration of the behavior-tree-driven pick-and-place pipeline:

In this video, the robot attempts to pick and place three objects, but struggles to pick up the mustard bottle.

Here, we can see the user prompting "the jello box," but Grounded SAM incorrectly detects the sugar box. The user invalidates the detection, causing the behavior tree to reset the perception pipeline and re-prompt for object detection.

Source code can be found on GitHub.