Robot's Decisions with Neural Network

Mar 3, 2006 - Using C++ Language to program the robot's controller. ▫ Testing the controller with Player + Gazebo. Simulator. ▫ If successfully testing under ...
223KB taille 3 téléchargements 335 vues
Friday, March 03, 2006

Robot’s Decisions with a Neural Network By Jirakhom Ruttanavakul University of Nevada, Reno CS 790R

Motivation

?

Challenges „ What do we need to consider if we deploy a robot

to harvest in the field? „ How can we control the robot to „

Do the right thing? „ Pick the right choice? „ Go the right way?

Implementation „ Rule-Based System (A new kind of science) „

Control next stage of the robot after a particular task is completed.

„ Inhibition – Activation (Animal’s Pattern) „

Making a decision

„ Reaction – Diffusion (Waves) „

(Not yet in this project)

„ Neural Network (Training or Learning) „

Control the level of Inhibition - Activation

Simulation „ Using C++ Language to program the robot’s

controller „ Testing the controller with Player + Gazebo Simulator „ If successfully testing under simulation, port the controller to the real robot and test again. „ Download simulation from „

http://playerstage.sourceforge.net/

Gazebo Screen Shots

Status „ Need Further Discussion with Professor such as

scope and more detail.

Network Template Object 00 Camera

Move Object 01

Speed Command

Collecting Fiducial

Object 02

Turn

Clean-UP Object 03

Turn Command

Gripper

Object 04

Charging

Close Grip

Laser Gripper Command

Object 05 Open Grip

Sonar Object 06

Battery

Robot

Example of Complete Network Move Camera Object 00

Collecting

Turn

Speed Command

Fiducial Move Clean-UP Gripper

Object 01

Charging

Turn

Turn Command

Robot

Close Grip Laser Object 01

Move Gripper Command

Sonar Turn

Battery

Close Grip

Example of network created when the robot receives the image of two object_01s and non others through camera

Comments & Questions