The Equipment Outline of the Humanoid Robot RO-PE and the Self-limitation Calculation in RoboCup.

Uploaded on:
The robot is not told its introductory stance, but rather needs to focus it from the earliest starting point ... a very much confined robot is teleported to some other. position without ...
Slide 1

The Hardware Design of the Humanoid Robot RO-PE and the Self-limitation Algorithm in RoboCup Tian Bo Control and Mechatronics Lab Mechanical Engineering 20 Feb 2009 SMC

Slide 2

RoboCup and Team RO-PE RoboCup TM is a universal joint venture to advance counterfeit consciousness and mechanical technology. Group RO-PE RO-PE (RObot for Personal Entertainment) is a progression of little size humanoid robots created by the Legged Locomotion Group

Slide 3


Slide 4

Design of the Robot

Slide 5

Hierarchy of the System

Slide 6

Self-limitation in RoboCup Global restriction issue The robot is not told its underlying stance, but rather needs to decide it from the earliest starting point

Slide 7

Self-confinement in RoboCup Global confinement issue Kidnapped robot issue - an all around limited robot is teleported to some other position without being told - The grabbed robot issue is regularly used to test a robot\'s capacity to recoup independently frame cataclysmic limitation disappointments

Slide 8

Self-confinement in RoboCup Global confinement issue Kidnapped robot issue Other troubles in the humanoid soccer situation - The field of perspective is constrained, because of the human-like sensor. - Noisy recognitions and uproarious odometry. - Computational assets are constrained. In any case, information should be handled continuously.

Slide 9

What is Particle Filter Belongs to the group of Bayesian Filters (Bayes Filter,Kalman Filter… ) Bayesian channel procedures give an intense factual device to oversee estimation instability Based on the information of past state, Bayesian channel probabilistically assesses a dynamic framework\'s state from boisterous environment.

Slide 10

What is Particle Filter Particle channels speak to convictions by an arrangements of tests, or particles It is a probabilistic methodology, in which the present area of the demonstrated as a thickness of the particles. Every molecule can be seen as the speculations of the robot being situated at this stance.

Slide 11

What is Particle Filter The primary goal of molecule sifting is to "track" a variable of enthusiasm as it advances after some time, commonly with a non-Gaussian and possibly multi-modular pdf The molecule channel calculation is recursive in nature and works in two stages: expectation and overhaul

Slide 12

Particle Filter Localization Move every one of the particles as per the movement model of the past activity of the robot More down to earth part Determine the probabilities q i in light of perception model (genuine "trap") Resampling

Slide 13

Algorithm particle_filter ( S t-1 , u t-1 z t ): For Generate new specimens Sample list j(i) from the discrete circulation given by w t-1 Sample from utilizing and Compute significance weight Update standardization component Insert For Normalize weights Particle Filter Algorithm (Probabilistic Robotics, C4 P98)

Slide 14

Particle Filter for Self-Localization Loop initializeParticles();/p[i] (x, y, theta, w) While(sensor reset != 1){ motionModel(); sensorModel();{ updateWeight(); } resampling(); yield(); }

Slide 15


Slide 16

Motion Model This is the forecast part The molecule channel for self-confinement appraises the robot " s posture Odometry-based Method Take x for instance: p[m].x = p[m].x + deltaX *(1+gaussian)

Slide 17

Motion Model Simplified Leg Model Step 1 hip_yaw = 0

Slide 18

Motion Model Simplified Leg Model Step 2 hip_yaw = θ

Slide 19

Motion Model We do limitation when the left leg simply touch the ground This odometry gets the information from the movement praise sent to servo, it won\'t be influenced by the control signal. It can be more exact if the servo can criticism its position.

Slide 20

Error for the Motion Model(1) with the strides expanded, the mistake increment. The biggest blunder is 25%, happens at the fourteenth step.

Slide 21

Error for the Motion Model(2) Like strolling movement, the genuine separation has a direct association with the quantity of steps, so we can accomplish better results through enhance our model or make revision. Back to Particle Filter

Slide 22

Sensor Model This is the overhaul part In the entire field, we just utilize the two objectives and two shafts for self-limitation. The world model is known. We just take the point from the historic point to the front of robot into thought

Slide 23

Sensor Model We are just utilizing the wide edge camera for milestone acknowledgment, the data we can digest from the camera is constrained.

Slide 24

Sensor Model Once a point of interest is seen by the robot, the capacity sensorModel() will be executed. The weight for each molecule will be upgraded as needs be. In the event that few historic points are seen on the double, the weight will be

Slide 25

Sensor Model We can get the expectedTheta through the position and introduction of the molecule and the world model if(blue_goal_found){/*we can get the percievedTheta from camera, the coordination of the milestone on the picture, and the position of the panning servo of the head */ updateWeight(blue_goal)}

Slide 26

Sensor Model Update Weight deltaTheta = fabs ( expectedTheta – perceivedTheta) ; conviction = dissemination ( deltaTheta ); p[i].weight = p[i].weight * conviction Normalize(p[i].weight); Distribution Policy Now we are utilizing Gaussian conveyance.

Slide 27

Resampling The least complex strategy for resampling is to choose every molecule with a likelihood equivalent to its weight. Select with Replacement Linear time Resampling Resampling by Liu et al.

Slide 28

Resampling (A Particle Filter Tutorial for Mobile Robot Localization TR-CIM-04-02)

Slide 29

Final Estimation Finding the Largest Cluster Give the best result yet computational costly Calculating the Average May influence by the distant particles Best Weight Fastest approach to give the outcome, appropriate for the constant framework

Slide 30

Future work Find out the condition to make sensor resetting , or else here and there the molecule will join to a false point and can\'t recoup. Counting the separation data in sensor model. Attempt new resampling and weightUpdate calculation.

Slide 34

Thank you

View more...