Sensor Guided Behaviors for a Dynamical Hexapod Ro

Uploaded from authorPOINTLite
Views:
 
Category: Entertainment
     
 

Presentation Description

No description available.

Comments

By: beny.siri (15 month(s) ago)

pls allow me to dwl this ppt

Presentation Transcript

Slide1: 

UPenn-CMU Meeting Sensor-Guided Behaviors for a Dynamical Hexapod Robot Sarjoun Skaff Carnegie Mellon University Friday 8 August 2003

Introduction - RHex: 

Introduction - RHex Characteristics High-speed, high-energy mobility Aptitude to overcome obstacles Limitations Challenging to operate High transient dynamics

Sensor-Guided Behaviors: 

Sensor-Guided Behaviors State Estimation for Dynamical Robots Sensor-Guided Behaviors Inertial Guidance Inertia Sensing

Related Work: 

Related Work Low Frequency Transients High Magnitude Transients Periodical Transients Complex Dynamics Low Magnitude Transients Autonomous Navigation Sensor Based Behavior [Amidi 98] [Wang 02] [Lu 00] [Del Rios 99]

Inertia-Guided Behavior - Formulation: 

Goal Use on-board gyroscope to maintain heading: Alleviate operator workload Stabilize gait at running speed Assumption - Steering control - Unicycle motion model: Sensor Gyroscope measures angular rate at 300Hz Inertia-Guided Behavior - Formulation vf u x y

Inertia-Guided Behavior - Approach: 

Inertia-Guided Behavior - Approach Motion model Controller Result Success up to jogging Motion model (running speed): Controller Result Model accurate enough With Inertial Guidance Without Inertial Guidance

Sensor-Guided Behaviors: 

Sensor-Guided Behaviors State Estimation for Dynamical Robots Sensor-Guided Behaviors Inertial Guidance Inertia Sensing

Vision-Guided Behavior – Formulation: 

Vision-Guided Behavior – Formulation Goal Use on-board camera to follow line and minimize operator control for this task Sensors Sony DFW-V300 at 30Hz Gyroscope Software (David Maiwand) Line extracted through color segmentation

Vision-Guided Behavior – Approach (1): 

Motion Model Controller Result Success up to jogging speed Vision-Guided Behavior – Approach (1) Linearize Re-write

Vision-Guided Behavior – Approach (2): 

Alternative Controller for Running Incorporate rate of rotation from gyroscope Result 30% success rate (failures mainly due to transient dynamics). Lessons Learned Model simplification can enable successful control of complex machines - RHex’s model changes in structure with speed (1st2ndorder) Vision-Guided Behavior – Approach (2)

Sensor-Guided Behaviors: 

Sensor-Guided Behaviors State Estimation for Dynamical Robots Sensor-Guided Behaviors Inertial Guidance Inertia Sensing

Localization(SLAM) - Definitions: 

Localization(SLAM) - Definitions State includes robot and landmarks

Localization(SLAM) - Illustration: 

Localization(SLAM) - Illustration 1. Predict 2. Observe 4. Update k k+1 1 2 3 4 5 3. Associate Data Gain reflects relative confidence in process and measurement accuracy Kalman Filter: Predict  Observe  Update

Coverage Setup: 

Sensor Camera Data Collected Range and Bearing Motion Model Unicycle Coverage Controller Back and Forth Area Sweeping Coverage Setup

Experiment 1 – Vision Sensing: 

Experiment 1 – Vision Sensing Two Reasons for Failure When Turning Motion model accuracy deteriorates with transient dynamics New landmarks seen briefly have uncertain location

Experiment 2 – Vision & Inertia Fusion: 

Experiment 2 – Vision & Inertia Fusion Gyroscope complements vision when Vision fails to capture landmarks Motion model accuracy deteriorates

Conclusion: 

Conclusion Sensor-based behavior enables automation of tasks of increasing complexity Simplified models can be sufficient for control and state estimation Performance of control and state estimation depends on accuracy of motion model Fusion of Camera and IMU data compensates for the degradation of visual information and motion models

Contributors: 

Contributors Al Rizzi David Maiwand Howie Choset

Appendix – Measurement Model: 

Appendix – Measurement Model Problem Measurement expressed in range and bearing, not in (x, y) coordinates Solution Linearize observation ?

Appendix – Kalman Filter Equations: 

Appendix – Kalman Filter Equations System Predict Observe Update Linear System Sensor Space Mean Spread in Work Space Spread in Sensor Space