THESIS
2021
1 online resource (xv, 90 pages) : illustrations (some color)
Abstract
When moving objects that are too bulky or heavy to be grasped or lifted, robotic manipulation
can benefit from the object’s interaction with a support surface and its natural
dynamics under gravity. This thesis presents the method of robotic rock-and-walk manipulation
for dynamic, non-prehensile and underactuated object transport. The object, which is
in contact with the support surface, is basically manipulated to rock from side to side about
the contact point periodically by the robot system. In the meantime, the passive dynamics
due to gravity enables the object to roll along a zigzag path that leads to a forward walk.
This work is motivated by an interesting question in archaeology, how the giant statues of
Easter Island (known as “moai”) were transported several hundred years ago,...[
Read more ]
When moving objects that are too bulky or heavy to be grasped or lifted, robotic manipulation
can benefit from the object’s interaction with a support surface and its natural
dynamics under gravity. This thesis presents the method of robotic rock-and-walk manipulation
for dynamic, non-prehensile and underactuated object transport. The object, which is
in contact with the support surface, is basically manipulated to rock from side to side about
the contact point periodically by the robot system. In the meantime, the passive dynamics
due to gravity enables the object to roll along a zigzag path that leads to a forward walk.
This work is motivated by an interesting question in archaeology, how the giant statues of
Easter Island (known as “moai”) were transported several hundred years ago, and a recent
demonstration performed by archaeologists that it is possible to walk the statue by repetitive
rocking.
In this thesis, we first study the passive, dynamic behavior of a representative object
model on a support surface and discuss its availability in relation to the geometric and mass properties of the object. Next, we separately develop model-based and learning-based control
methods to realize dynamic rolling gait for object transport. In the model-based approach,
we devise a feedback control strategy for steady object transport through effective regulation
of its energy and posture by rock-and-walk manipulation. In the learning-based approach,
we acquire a similar manipulation capability through reinforcement learning in a dynamic
simulation environment that features the object and the support surface. An extensive set of
experiments demonstrate the viability and practicality of our approaches in diverse settings in
which the robots interact with the object in a non-prehensile manner. These include caging-based
single-robot manipulation and cable-driven dual-robot manipulation using manipulator
arms and aerial robots.
Post a Comment