Page 1

xiAac�: Autonomous Emotional Robot

is Learning Soccer

Mahdi Ghayoumi1, 5, Mohammad Shayganfar1, 5, Payam Porkar1, 5, Esmaeel Zamani2, 5, Peyman Soltani2, 5, Seyed Abdoreza Pishvaei3, 5,

Ali Ghasemnezhad4, 5, Masood Teymoori4, 5, Seyed Nooredin Rasi Hashemi4, 5,

1Artificial Intelligence & Robotic Engineering Department, 2Systems and Control Engineering Department,

3Computer Architecture Engineering Department, Science and Research Unit of Tehran Azad University - Prof. M.H. Korayem,

Tehran, Iran 4Mechanical Engineering Department,

Central Unit of Tehran Azad University - Dr. A. Adamian Tehran, Iran

5Humanoid Robots Branch, Aryaak� Robotic Group, Aryaak� Research Center,

14817-37791 Tehran, Iran Humanoid@Aryaak.com http://www.Aryaak.com

Abstract. Our new project, xiAac� is fully autonomous robot with 26 degree of free-dom, stereovision network camera, 3-axis acceleration, 3-axis gyro sensors, force and weight sensor that detect angular velocity. We use intelligent servo and stepper actuator for robot motion. Motion analysis was estimated with MSC.ADAMS� 2005 and 3D CAD map was designed with CATIA� V5 R16.

1 Introduction

Aryaak� Humanoid Team is the one of Aryaak� Robotic Group Teams that tries to participate at RoboCup 2006. It is in cooperation with various departments of Tehran Azad University. Now our previous robot with Firatelloid name is under redesigning and develop-ing for Humanoid soccer league in RoboCup 2006.

Page 2

2 Specifications of xiAac�

According to Table 1, xiAac� were designed with 26 degrees of freedom and 3 types of intelligent actuator. All of details were based on Aluminum alloy but it is now under reprocessing for composite materials. This method create better environ-ment for PC Table and microcontrollers. The robot weight is 10.6 kg and height, weight and depth are 915, 320 and 10 mm.

2.1 Mechanical and Motion Specification

Table 1: DOF Specifications

Joints Number of DOF Actuator Type Neck 2 DOF Type 1 Waist 2 DOF Type 2 Shoulders 3 DOF + 3 DOF Type 2 + Type3 Elbows 1 DOF + 1 DOF Type 2 Wrists 1 DOF + 1 DOF Type 2 Hips 3 DOF + 3 DOF Type 3 Knees 1 DOF + 1 DOF Type 3 Ankles 2 DOF + 2 DOF Type 3

Total 26 DOF

Table 2: Specifications of Actuators

Data Type 1 Type 2 Type 3 Manufacturer Bohler� Motor Maxon� Motor Maxon� Motor Type Intelligent Stepper Intelligent Servo Intelligent Servo Model ---------------- RE-Max 24 Re-Max 29

Motor Size 29.2 mm × 22 mm 31.9 mm × 24 mm 44.7 mm × 29 mm Gear Size 23.1 mm × 25 mm 26.4 mm × 26 mm 38.2 mm × 32 mm Torque 4.2 nm 5.9 nm 6.3 nm Speed 0.217. rpm 0.198 rpm 0.226 rpm

Page 3

Figurate 1: Different views of 3D Graphical CAD, Designed with CATIA� V5 R16

Page 4

Motion Control

In this project, motion control of mobile robot by hybrid control system and for modeling of system inverse Kinematics is base for control system used both of feed forward and feed back control .First the developed control scheme can adaptively estimate the dynamics of the robot using neuro-fuzzy method, and based on error dynamics of a feedback controller, a learning rule for updating the connection weights of neural control system .second For predict state of robot, using Model Predictive Controllers (MPC).

Figure 2: Knee Motor Power Consumption Analysis with MSC.ADAMS� 12

.

Page 5

Figure 3: Compare of Physical Data Analysis and Virtual One of Ankle with MSC.ADAMS� 2005 (Measurement Positions in X&Z Axes)

Page 6

2.2 Electrical Specifications

Hardware designing is based on parallel processing that allows better and modu-lar control. For modular controlling, in this case hardware design fall in master - slave technique. A table PC is master and controls all over the system and leads microcontrollers. Microcontrollers are in slave mode that communicates by master with a bridge interface. All slave part goes in two types. Microcontroller that get information from sen-sors and use for input and feedback data to master, and microcontrollers that control motors that order to driver and use for output instructions. These two controller classes work symmetrical. Input information from input controller class given to master for movement processing and new rules come for output microcontroller class to performed action. There are some interface between microcontrollers, table PC, sensors and motor driver.

Table PC and Microcontrollers

Here, Table PC works like robot's brain. Complicated and heavy processing ac-complish on it. High level input and output directly connect on table PC for data manipulation. It also is able to communicate with other computer. This packet is similar to small PC that installs on robot and controls all parts. Order generate with master and distributed to microcontrollers for accomplish. Recommended CPU for this robot is AMD� Geode� GX533 and Microcon-trollers are 89C52, manufactured by ATMEL� under INTEL� Corporation license with 24 MHZ speed. Selection parameter for choice microcontroller is capability of bit manipulation and serial communication. Firmware was written with C++ pro-gramming language.

Sensors

We use force and weight sensor to load balancing and have better movement. Force sensors are on resistance type and control with analog part like UP-AMP and tune for work properly and also use vibration sensor for correct errors on movement. These are the most user-friendly system available for controlling and sensing the environment from machines. Sensors are manufactured by PHIDGETS� Company.

Page 7

Figure 4: The Block Diagram of Network Camera

Camera

Camera that selected for this project is network camera. A network camera can be described as a camera and computer combined into one unit. It connects directly to the network as any other network device. A network camera has its own IP address and built-in computing functions to handle network communication. Everything needed for viewing images over the network is built into the unit. A network camera has built-in software for a Web server, FTP server, FTP client and e-mail client. Other features include alarm input and relay output functions. More advanced network cameras can also be equipped with many other value-added op-tions such as motion detection and an analog video output. The lens of the camera focuses the image onto the image sensor (CCD/CMOS). Before reaching the image sensor, the images pass through the optical filter, which removes any infrared (IR) light so that the "correct" colors will be displayed. (In day/night cameras, this IR-cut filter is removable to provide high quality black & white video during nighttime conditions.) With a built-in Web server, a network camera does not need a direct connection to a PC or any other hardware or software to capture and transfer images. It operates as a stand-alone unit and requires only a connection to the network.

Camera Controlling Software

We have implemented the camera controlling software with C++ language. Soft-ware includes two parts of Video Capturing and also functionality control of the camera. The first part uses a dedicated ActiveX which properly is used to preview captured video streams. Second part helps to implement a bi-directional connection between operator and the camera. It means that the operator can control existent capabilities like Pan, Tilt and Zoom, also some other image features like its white-balance, color-level, and transferring image compression beside any other features that can help us for a better quality of control. This part is implemented as a win32 dll file which included.

Page 8

2.3 Vision and Image Processing Specifications

Our vision software in robot league detects key objects based on their color (color matching) and estimates their coordinates in a real world (stereo vision). In this system a combination of some algorithms such as color matching, segmentation and stereo vision is applied. In this system some heuristic ideas are presented. These ideas are used to reduction of vision error. Of course these ideas are logical and often according to an AI method.

Figure 5: Original Image

Figure 6: Segmented Image (for other processing)

Page 9

3 Research approaches with xiAac�

3.1 Firatelloid Robot

Firatelloid, First Iranian Intelligent Humanoid robot, has been built through the years 1999-2001. It was the first experiment of building a human-like robot after working on different simple and competitive robots leading to various fields of inter-est for each member of the team. All hardware and software controlling systems were designed by fellows.

3.1.1 Local Controlling Interface

Audio/Video 2.4 GHz receiver Digital wireless serial data transceiver module to send proper user com-

mands to the robot and receive robot�s environmental status 8-bit u-controller to control all existing properties of this box Character LCD to show data/device status

Page 10

This device was designed to operate as an interface between the PC and the robot. All user commands were being sent through the LPT and translated into appropriate commands by the u-controller, and then these commands could be sent to the robot via the digital serial data transceiver. The PC also could receive any robot�s sensor-status by this module, and at the other hand user could understand the status of the interface device by using the installed character LCD independently, and as a main part, Video streams from the robot could be captured by an existing transceiver which was plugged to the PC via BNC connector.

3.1.2 Hardware Specifications

Master-Slave u-controller based parallel manipulators controlling architec-ture

Audio/Video 2.4 GHz transmitter Digital wireless serial data transceiver module to get proper user commands

and send environmental status of the robot Two CCD camera DC and stepper Motors Sensors Motor driver circuits Sensor interface circuits Sealed-lead-acid batteries

The robot could get all his commands through a wireless serial data transceiver module. This module was connected and controlled via a u-controller, and then re-ceived data could be analyzed with a master u-controller to determine which manipu-lator�s slave u-controller should get and run them as its own received commands. Since each independent manipulator could work together, we had a single u-controller for each as the architecture�s slaves. These slaves could get and functional-ize their commands and at the other hand the navigation of robot such as automatic obstacle (static and moving) detection could be handled through the analyzing of all receiving feedbacks, which made its navigation autonomous. IR (Infra-Red), Ultra-sound, FSR (Force Sensing Resistors), Microwave sensors all were mounted on this robot that made the navigation much more easier by detect-ing obstacles with pre-defined distance, moving obstacles in front of the robot, and moving objects behind obstacles. Robot�s CCDs made the navigation more practical, which was implemented in two non/autonomous modes. Some heuristic algorithms were taken into the job for better autonomous navigation.

Page 11

3.1.3 Software Specifications

The software was also can be divided into two main parts that was written in the below:

PC-based software

PC-based software which was implemented on win32 architecture by MFC Visual C++ with the following characteristics:

Multi-threaded Architecture Video Capturing Sub-system Image Processing Sub-system Speech Recognition Sub-system User-friendly GUI Robot Status Analyzing Sub-system Local Hardware Controlling Sub-system Voice-based Interaction Sub-system (Persian-English-German Languages)

Micro-based Software

Micro-based Software which was implemented by C51 language with the follow-ing characteristics:

Serial port interrupt handling External interrupt handling Timers interrupt handling Working with serial and parallel ports Working with ADCs Controlling stepper motor drivers Controlling DC-motors using PWM Controlling Character and graphical LCDs

Page 12

4. Deduction

In this TDP, we described our robot named xiAac� that has 26 degrees of free-dom, and several kinds of sensors to autonomously generate various motions. We also proposed a method to recognize environment and select appropriate behavior for humanoid robots based on sensor histories.

5. References

1. I. Witten and E. Frank, "Data Mining - Practical Machine Learning Tools and Techniques with JAVA Implementations -," Morgan Kaufman, 2000. 2. Steven H. Collins, Martijn Wisse, and Andy Ruina: A 3-D passive dynamic walking robot with two legs and knees. International Journal of Robotics Research, 20(7), pp. 607-615, 2001. 3. Tad McGeer: Passive dynamic walking. International Journal of Robotics Research, 9(2),

pp. 62-82, 1990.