Efficient Unicontact Grasping in Cluttered Scenes

Vishal Satish

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2021-112
May 14, 2021

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-112.pdf

Mechanical search -- finding and extracting a known target object from a cluttered environment -- is a key challenge in automating warehouse, home, retail, and industrial tasks. In this thesis we consider contexts where occluding objects are to remain untouched to minimize disruptions and avoid toppling. We assume a 6-axis robot has an RGBD camera and suction gripper mounted on its wrist, such that it can move to an approach vector along which the suction gripper can both be inserted to grasp the target object and then retracted to extract it. We formalize the problem of efficiently finding an approach vector and present AVPLUG: Approach Vector PLanning for Unicontact Grasping: an algorithm using a fast oct-tree occupancy model and Minkowski sum computation to maximize information gain. Experiments in simulation and with a physical Fetch robot suggest that AVPLUG finds an approach vector up to 10x faster than a baseline search policy.

Advisor: Ken Goldberg


BibTeX citation:

@mastersthesis{Satish:EECS-2021-112,
    Author = {Satish, Vishal},
    Editor = {Goldberg, Ken and Kanazawa, Angjoo},
    Title = {Efficient Unicontact Grasping in Cluttered Scenes},
    School = {EECS Department, University of California, Berkeley},
    Year = {2021},
    Month = {May},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-112.html},
    Number = {UCB/EECS-2021-112},
    Abstract = {Mechanical search -- finding and extracting a known target object from a cluttered environment -- is a key challenge in automating warehouse, home, retail, and industrial tasks.  In this thesis we consider contexts where occluding objects are to remain untouched to minimize disruptions and avoid toppling.  We assume a 6-axis robot has an RGBD camera and suction gripper mounted on its wrist, such that it can move to an approach vector along which the suction gripper can both be inserted to grasp the target object and then retracted to
extract it. We formalize the problem of efficiently finding an approach vector and present AVPLUG: Approach Vector PLanning for Unicontact Grasping: an algorithm using a fast oct-tree occupancy model and Minkowski sum computation to maximize information gain. Experiments in simulation and with a physical Fetch robot suggest that AVPLUG finds an approach vector up to 10x faster than a baseline search policy.}
}

EndNote citation:

%0 Thesis
%A Satish, Vishal
%E Goldberg, Ken
%E Kanazawa, Angjoo
%T Efficient Unicontact Grasping in Cluttered Scenes
%I EECS Department, University of California, Berkeley
%D 2021
%8 May 14
%@ UCB/EECS-2021-112
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-112.html
%F Satish:EECS-2021-112