Quadruped locomotion reference synthesis with central pattern generators tuned by evolutionary algorithms

Adak, Ömer Kemal (2013) Quadruped locomotion reference synthesis with central pattern generators tuned by evolutionary algorithms. [Thesis]

[thumbnail of ÖmerKemalAdak_463008.pdf] PDF
ÖmerKemalAdak_463008.pdf

Download (2MB)

Abstract

With the recent advances in sensing, actuating and communication tecnologies and in theory for control and navigation; mobile robotic platforms are seen more promising than ever. This is so for many fields ranging from search and rescue in earthquake sites to military applications. Autonomous or teleoperated land vehicles make a major class of these mobile platforms. Legged robots, with their potential virtues in obstacle avoidance and cross-country capabilities stand out for applications on rugged terrain. In the nature, there are a lot of examples where four-legged anatomy embraces both speed and climbing characteristics. This thesis is on the locomotion reference generation of quadruped robots. Reference generation plays a vital role for the success of the locomotion controller. It involves the timing of the steps and the selection of various spatial parameters. The generated references should be suitable to be followed. They should not be over-demanding to cause the robot fall by loosing its balance. Nature tells that the pattern of the steps, that is, the gait, also changes with the speed of locomotion. A well-planned reference generation algorithm should take gait transitions into account. Central Pattern Generators (CPG) are biologically-inspired tools for legged-robot locomotion reference generation. They represent one of the main stream quadruped robot locomotion synthesis approaches, along with Zero Moment Point (ZMP) based techniques and trial–and–error methods. CPGs stand out with their natural convenience for gait transitions. This is so because of the stable limit cycle behavior inhertent in their structure. However, the parameter selection and tuning of this type of reference generators is difficult. Often, trial–and–error iterations are employed to obtain suitable parameters. The background of complicated dynamics and difficulties in reference generation makes automatic tuning of CPGs an interesting area of research. A natural command for a legged robot is the speed of its locomotion. When considered from kinematics point of view, there is no unique set of walking parameters which yield a given desired speed. However, some of the solutions can be more suitable for a stable walk, whereas others may lead to instability and cause robot to fall. This thesis proposes a quadruped gait tuning method based on evolutionary methods. A velocity command is given as the input to the system. A CPG based reference generation method is employed. 3D full-dynamics locomotion simulations with a 16-degrees-of-freedom (DOF) quadruped robot model are performed to assess the fitness of artificial populations. The fitness is measured by three different cost functions. The first cost function measures the amount of support the simulated quadruped receives from torsional virtual springs and dampers opposing the changes in body orientation, whereas the second one is a measure of energy efficiency in the locomotion. The third cost function is a combination of the firs two. Tuning results with the three cost functions are obtained and compared. Cross-over and mutation mechanisms generate new populations. Simulation results verify the merits of the proposed reference generation and tuning method.
Item Type: Thesis
Uncontrolled Keywords: Quadruped robots. -- Locomotion reference generation. -- Central Pattern Generation. -- Gait tuning. -- Genetic algorithms. -- Dört bacaklı robotlar. -- Dört bacaklı yürüme referansı oluşturulması. -- Merkezi örüntü üretimi. -- Genetik algoritma.
Subjects: T Technology > TJ Mechanical engineering and machinery > TJ163.12 Mechatronics
Divisions: Faculty of Engineering and Natural Sciences > Academic programs > Mechatronics
Faculty of Engineering and Natural Sciences
Depositing User: IC-Cataloging
Date Deposited: 29 Sep 2014 11:53
Last Modified: 26 Apr 2022 10:02
URI: https://research.sabanciuniv.edu/id/eprint/24614

Actions (login required)

View Item
View Item