Tutorial Darwin2K

Voeg

Le téléchargement nécessite un accès à la bibliothèque YouScribe
Tout savoir sur nos offres

101 pages

English

Le téléchargement nécessite un accès à la bibliothèque YouScribe
Tout savoir sur nos offres

A propos
Informations
Extrait

Description

Informations

Publié par	Voeg
Nombre de lectures	16
Langue	English

Extrait

Darwin2K Tutorial
Chris Leger
The Robotics Institute
Carnegie Mellon University
Pittsburgh, Pennsylvania 15213
Last updated: 6/13/00
Copyright © 2000 Chris Leger.
All rights reserved.Darwin2K Tutorial
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Building Robot Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Using the Synthesizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5 Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3Darwin2K Tutorial
4Darwin2K Tutorial
Figure 1.1: Software architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Figure 2.1: Sample parameterized module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Figure 2.2: ThehollowTube module. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Figure 2.3: Component context and lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Figure 2.4: Component description for a gearbox. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Figure 2.5: Sample configuration and text description. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Figure 2.6: Configuration shown as modules and links. . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Figure 2.7: Configuration graph for a two-armed robot . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Figure 2.8:displayCfg GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 2.9: disp.p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5Darwin2K Tutorial
6Darwin2K Tutorial
Table 2.1: Component properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7Darwin2K Tutorial
8Darwin2K Tutorial
1 Introduction
1.1 What is Darwin2K?
Darwin2K is a software package for simulating, synthesizing, and optimizing ro-
bots. The motivation behind Darwin2K is to provide a set of software tools to make the
robot configuration process easier for the robot designer, and to allow better robots to be
designed. Traditionally, the configuration design process (or configuration synthesis) has
addressed the creation of the high-level form of the robot: whether it will be a manipula-
tor or mobile robot, whether the robot will have wheels or legs or some combination
thereof, and so on. After a particular configuration is selected, the dimensions, compo-
nents, and other specific properties are determined; this may be called parametric design.
In some cases, numerical analysis and optimization may be undertaken to determine the
best values for specific properties, though often values which simply satisfy a set of re-
quirements are selected by hand.
Darwin2K combines some aspects of the configuration and parametric design pro-
cesses, for two related reasons. First, the performance and behavior of a configuration of-
ten cannot be accurately predicted without specific parametric values. Second, Darwin2K
uses simulation to measure a robot’s performance, and simulation requires knowledge of
the robot’s configuration as well as its parametric properties. However, Darwin2K does
not address the entire configuration synthesis problem, nor the entire parametric optimi-
zation problem. Instead, it tackles the medium- to high-level parametric design and opti-
mization problem, and the low- to medium-level configuration synthesis problem. For
example, while Darwin2K can optimize a robot’s kinematic dimensions and actuator
choices, it will not generate a detailed design with information about cable routing, bolt
patterns, and so on. This lowest level of detail for a design can usually be generated from
a higher-level design without requiring many design iterations--that is, the higher-level
design will not usually require significant modifications to address the issues arising in
detailed design.
At the other end of the design spectrum, Darwin2K will not make the highest level
of design decisions such as deciding whether a flying, swimming, walking, or rolling ro-
bot is best suited for a task. This decision process involves envisioning a task description
and accompanying motion plans, which must then be embodied in the choice of robot
controller, tool trajectories, and so on, before a configuration’s performance can be quan-
titatively evaluated. The designer can, however, use Darwin2K to generate reasonably
detailed, well-optimized designs for multiple categories (e.g. flying or walking) of config-
urations, allowing the designer to make an informed choice about configuration issues.
Simulating a robot requires the designer to provide a description of the task, including
properties such as trajectories, controllers, payload models; if the designer can provide
these for multiple classes of configurations, then Darwin2K can be used to synthesize and
optimize robots for each general class of configurations.
Introduction 9Darwin2K Tutorial
task
specification configurations
Synthesizer
performance data Evaluator
task-specific task-specific
library library
Figure 1.1: Software architecture
Darwin2K consists of two main programs: the synthesizer (or Population Manager),
and the evaluator. The evaluator includes general-purpose libraries for simulating
and controlling robots, while dynamic libraries allow the designer to provided task-
specific modules, task descriptions, controllers, and simulation code. Many
evaluator processes can be distributed across multiple computers or processors to
reduce overall runtime.
1.2 Software overview
Darwin2K contains several programs and a suite of robot simulation libraries. The
two main programs are the synthesizer (or Population Manager (PM)) and the evaluator
(see Figure 1.1). The synthesizer generates robot designs and sends them to the evaluator,
which simulates the robot and returns performance measurements to the synthesizer. In
practice, many evaluator processes can be distributed across a network of computers (or
among multiple CPUs in a multiprocessor computer) since evaluating designs is the most
computationally expensive part of the synthesis process. The evaluator program makes
use of Darwin2K’s libraries of robot parts and simulation and control algorithms. Some
of the features included in these libraries are:
• kinematic simulation
• dynamic simulation
• Jacobian-based trajectory following
• PID control
• estimating link deﬂection
• computing joint torques
• collision detection
The libraries also contain a few dozen parameterized modules, which represent robot parts
such as links, joints, end effectors, and mobile bases. These modules are the building
blocks from which the designer and the synthesizer construct robots.
While a wide range of design problems can be addressed with the capabilities and
modules included in Darwin2K, there will always be tasks for which special-purpose con-
10 Software overview