Dexterous robotic manipulation: Batting and cutting
Date
2021-12
Authors
Mu, Xiaoqian
Major Professor
Advisor
Jia, Yan-Bin
Fernandez-Baca, David
Tian, Jin
Oliver, James
Stoytchev, Alexander
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Altmetrics
Abstract
Dexterous robotic manipulation aims at achieving human-like agility and dexterity. In this thesis, we focus on two manipulation tasks: batting and cutting. Both tasks require the robot to react in an accurate, fast, and smooth way as a human does.
Batting a flying object to a target is a skillful maneuver. In robotic manipulation, precision batting remains one of the most challenging tasks as it leverages computer vision, modeling, motion planning, and robotic control, etc. We investigate targeted batting by a two-degree-of-freedom robotic arm, assuming all its movements to take place in one vertical plane. With the object’s pre-batting configuration predicted, the robot-object impact can be solved under Coulomb friction and energy-based restitution. Motion planning for the robot is conducted under constraints from impact dynamics, projectile flight mechanics, robot kinematics, etc. Planning is executed repeatedly upon every new estimate of the object’s motion. Meanwhile, the robot keeps adjusting its motion toward the final batting configuration. Experiments with different objects have claimed better performance than a human without training.
Skills of cutting natural foods are important for the robot looking to play a bigger role in kitchen assistance. In addition to achieving material fracture, there is an objective of executing fast and smooth motions of the kitchen knife, which in the process performs work to overcome material toughness, act against friction, and generate deformation on the object. To imitate human-like cutting, we decompose the cutting action into three phases: pressing, touching, and slicing. Position, impedance, and force controls are applied, either separately or jointly, in the three phases to cope with evolving contacts between the knife and the object or between the knife and the cutting board. In pressing, data acquired by a force/torque sensor can be used to estimate the object's properties such as Poisson's ratio, fracture toughness, and the product of the coefficient of blade-object friction and pressure distribution. These properties are all specific to the object and vary with its freshness and the portion undergoing fracture. The estimated property values can be applied in knife trajectory optimization in the same phase of pressing. Moreover, they can also be used for control purposes in the phase of slicing. Experiments conducted on several types of fruits and vegetables have exhibited natural cutting movements like those performed by a human hand. Cutting along the planned “optimal” trajectories has claimed less work than along those with constant knife moving directions.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
dissertation