controlling agents with natural language jared allen 2005 university of arkansas eia agent team

11
Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Upload: dorothy-bradford

Post on 29-Jan-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Controlling Agents with Natural Language

Jared Allen2005

University of Arkansas EiA Agent team

Page 2: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Background

● Increasing dependency upon machines- Grocery stores, household, etc.

● Complexity of devices- Examples: Thermostat, TV remote controls

● EiA vision- Everything is Alive

Page 3: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Natural Language

● Objective: An intuitive means of communicating and controlling electronic devices.

● Benefits- ease of use, computational power, limited hardware

● Limitations- implementation viability, scope of language

Page 4: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Menu-Based Natural Language Interface (MBNLI)

● Format- cascading menus, guided completion dialogue,

absolute grammar coverage

● Advantages- avoids under-shooting or over-shooting the allowable

language, retains ease of use

Page 5: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

MBNLI

Page 6: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Components

● Grammar: The set of allowable input commands.

● Parser: The software component that keeps track of instruction progress, future options, and output to the real-world device.

● Interface: The visual element of the system that guides user input.

Page 7: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Input/Output

● Input: Grammar, Command String, Translation Definitions

● Output: Machine Instruction (for our purposes, this will be in XML)

Page 8: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Current Implementation

AmigoBots

Page 9: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Robot Functionality

● Equipment- Sonar range devices, wireless internet adaptor,

gripper

● Functionality- Move, turn, grip, release, check sonar, etc.

Page 10: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

Example

“Robot, move forward 10 feet.”

<REQUEST> <CONTENTS>

<COMMAND><MOTION_COMMAND>

<COMMAND_TYPE> <![CDATA[move]]> </COMMAND_TYPE><COMMAND_ARGUMENT>

<ARGUMENT_TYPE> <![CDATA[double]]> </ARGUMENT_TYPE>

<ARGUMENT_VALUE> <![CDATA[10.0]]> </ARGUMENT_VALUE>

</COMMAND_ARGUMENT></MOTION_COMMAND>

</COMMAND> </CONTENTS> </REQUEST>

Page 11: Controlling Agents with Natural Language Jared Allen 2005 University of Arkansas EiA Agent team

For the future...

● Larger collection of grammars

● Cooperative grammars

● Intelligent Agents (“Smart Wrappers”)

● Remote access