ADAPTIVE ROBOT TRAINING EXPLORATIONS IN SENSORLESS MANIPULATION
Date
1990-06-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Several different techniques for training robots exist. The simplest
and most common is leading or guiding, but this method is limited and
inflexible. Textual or explicit programming has enjoyed many advances
in the last decade and current research versions of explicit systems
are very powerful. But explicit programming is complex and requires
the skills of specially trained programmers. Mixed systems attempt to
capitalize on the benefits of guiding and programming to simplify the
training process. This union is particularly applicable to training
sensorless robots in which the robot is used as a measuring device to specify
task locations. But existing mixed systems exhibit a poorly designed
interface, thus creating problems in the way the programming and guiding
processes interact.
This thesis presents the design and implementation of a prototypical mixed
system that improves not only the programming and guiding components,
but also their interface. The improvements are embodied in ART, an
Adaptive Robot Trainer. ART's development involved an analysis of
mixed systems and assembly tasks that resulted in an effective representation
of task state. The representation led to the design of ART's programming
language which automates much of the program-guiding interaction. ART's
syntax allows the programmer to express assembly operations and
object-feature relationships in a natural way while providing the
system with the information necessary to maintain task state. The
representation also enables guiding error corrections, flexibility
in the guiding protocol, and the generation of meaningful messages to prompt
operator actions.
Description
Keywords
Computer Science