Sensor and task planning in robotic assembly
Abstract
An intelligent robotic assembly must be able to determine the initial state of its work cell and create complete assembly plans given as input a high level description of assembly goals, geometric models of the components of the assembly, and a description of the capabilities of the work cell (including the robot and the sensory system). These assembly plans should include manipulations to achieve the assembly goals, sensing operations to verify the success of the manipulations, and recovery plans to be invoked when assembly errors occur. To determine the initial state of the work cell, we have developed an approach to planning sensing strategies dynamically. Sensing operations are proposed and evaluated in terms of their effect on the ambiguity in the current set of hypotheses about the objects in the work cell. When the resulting ambiguity is sufficiently small, the proposed sensing operation is selected for application. The Dempster-Shafer formalism for representing uncertainty is used to assess belief in individual hypotheses. To measure the ambiguity in a set of hypotheses, we apply the concept of entropy from information theory. The number of sensing operations which are considered is limited by grouping together equivalent sensing operations using a data structure which is based on the aspect graph. To create assembly plans, we have developed SPAR, a planning system which reasons about high level operational goals, geometric goals and uncertainty-reduction goals. Operational planning is done using a nonlinear, constraint posting planner. Geometric goals are satisfied by constraining the execution of actions in the plan, or, if the geometric configuration of the world prevents this, by introducing new actions into the plan. When the uncertainty in the world description exceeds that specified by uncertainty-reduction goals, sensing operations or manipulations are introduced in an attempt to reduce that uncertainty to acceptable levels. If this fails, the plan is augmented with verification sensing operations and local error recovery plans.
Degree
Ph.D.
Advisors
Kak, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.