logo SBA

ETD

Digital archive of theses discussed at the University of Pisa

 

Thesis etd-06152021-100244


Thesis type
Tesi di dottorato di ricerca
Author
GABELLIERI, CHIARA
URN
etd-06152021-100244
Thesis title
THE ROLE OF INTERACTION FORCES IN ROBOTIC MANIPULATION FOR LOGISTICS: A SPECIAL FOCUS ON DEPALLETIZING AND OBJECT DELIVERY
Academic discipline
ING-INF/04
Course of study
INGEGNERIA DELL'INFORMAZIONE
Supervisors
tutor Prof.ssa Pallottino, Lucia
tutor Prof. Bicchi, Antonio
Keywords
  • Depalletizing
  • Logistics Automation
  • Robotics
Graduation session start date
02/07/2021
Availability
Withheld
Release date
02/07/2024
Summary
THAT the passive compliance embedded in their mechanical design and the active
compliance conferred by their control laws have been allowing industrial robots
out of the cages in which they were traditionally constrained is acknowledged.
What the robots can do once out of their cages is an unfolding story that inspires this
Thesis.
Robotics is trying to respond to new industrial needs, such as high flexibility required
in warehouses, due especially to e-commerce, the necessity of lifting human
workers from the harsh working conditions that new efficiency standards impose, and
reducing the environmental impact of our activities. To do so, robots need to move
in uncertain and changing environments, sometimes shared with humans, and in general
not suitable for the old paradigms of high-velocity and high-precision predefined
movements. Physical contacts with the environment are not to be confined at all costs
anymore, but they become a useful means for adapting to an uncertain environment,
thus enhancing the robot capabilities.
The focus of this Thesis is the study of the interaction forces between the robot and
its surroundings, exploiting the role played by those forces in enabling novel logistic
applications. The topics range from depalletizing and object picking to object delivery.
First, this Thesis presents a strategy for autonomous pallet unwrapping, namely the
removal of the plastic film enclosing the parcels stacked on pallets. This is a key procedure
of the intralogistic flow, the automation of which has not yet received much
attention. Currently, unwrapping is performed mainly by human operators, due also to
the complexity of its planning and control phases. The unwrapping robot considered in
this Thesis is composed of a robotic manipulator equipped with a custom cutting end-effector.
A planning method aimed to ensure a successful task execution even when
the film profile is uncertain and irregularly shaped is presented. The proposed planning
strategy leverages information regarding the robot collisions with the environment. Experimental
results are presented to test the method.
After unwrapping the pallets, the underlying items become accessible for manipulation.
This Thesis also studies a planner that enables a novel depalletizing strategy for
the depalletizer robot WRAPP-up. The envisioned planner, exploiting contact force information, allows the dual-arm depalletizer to handle a large variety of goods even with
non-perfectly known positions. Results of experiments conducted on different items
provided by a food-delivery company are reported.
Besides bi-manual depalletizing, grasping smaller objects with a single end-effector
is a relevant task also in logistics and one that WRAPP-up can accomplish, too. The
problem of grasping a single object with an ad-hoc end-effector has been extensively
studied in the literature both from theoretical and experimental viewpoints but it still
represents an open problem, especially when no prior model of the objects is provided.
This Thesis presents a method for object picking, particularly suited for soft hands,
that can be used to grasp previously unseen objects. The major contribution is a data-driven
planner that generates suitable grasps by relying on a reduced database of grasps
performed by a skilled operator using the robotic hand. The basic idea is to exploit a
skilled human performing experiments using the robotic hand to grasp only a set of
basic shapes instead of general objects, dramatically reducing the number of trials.
The approach is then generalized to grasp unknown objects by relying on state-of-the-art decomposition algorithms that allow approximating an object with such basic
shapes. The method has been tested with the PISA/IIT SoftHand mounted on a Panda
robot and has shown a success percentage of 86.7% over 105 grasps on 21 previously
unseen objects.
Physical interaction capabilities can be entrusted also to aerial robots, enabling aerial
object manipulation. This may have rather interesting applications in logistics, especially
for last-mile delivery in the freight sector, allowing for a schedule not affected
by the unpredictable traffic jam, as in the case of road transportation. Cooperative approaches
to aerial object manipulation have been widely studied in the literature since
they may increase manipulation capabilities and would allow overcoming payload limitations.
However, communication among the robots, required for their coordination,
typically increases the system complexity and represents a possible source of low performance
and stability issues due to data loss, corruption, and delays.
This Thesis presents a method for the cooperative manipulation of cable-suspended
objects not requiring any explicit communication among the robots. Instead, the coordination
is enabled by an implicit form of communication that relies on contact-force
sensing. The method exploits a leader-follower scheme on admittance-controlled multi-rotors.
First, two robots and a beam load are considered. The stability and passivity of
the controlled system are discussed, as well as the effects of parametric uncertainties.
The specific role of the internal force induced on the object by the robots through the
cables is highlighted. An extension of the method to more than two robots is presented.
Eventually, the theoretical analysis is validated through numerical simulations, and the
results of preliminary experiments on two robots are presented as well.
File