Обсуждение участника:SAMSON

ПЕРЕВОДИМ (DESIGN AND CONTROL OF A ROBOTIC LEG WITH BRAIDED PNEUMATIC ACTUATORS)

Оглавление

Благодарности

Аннотация

Глава I. ВВЕДЕНИЕ

1.1 Background

1.2 Purpose and Overview

Глава II. Braided Pneumatic Actuators

2.1 Physical Characteristics

2.2 Geometric and Static Model

2.3 Static Model Verification

2.4 Dynamic Model

Глава III. Simulation

3.1 Simulation Overview

3.2 Equations of Motion

3.3 Valve Model

3.4 Dynamic Model Verification

Глава IV. Robot Hardware

4.1 System Overview

4.2 Leg Design

4.3 Valves

4.4 Force Sensors

4.5 Angle Sensors

Глава V. Control

5.1 Control Architecture and Control Laws

5.2 Control Program

5.3 Inverse Kinematics

Глава VI. Results and Discussion

6.1 Desired Walking Behavior

6.2 Tuning

6.3 Walking Results

6.4 Robot Limitations

6.5 Derivative Control

Глава VII. Conclusion

7.1 It Walks!

7.2 Semi-Observed Speculation

7.3 Future work

---

Приложение A: Simulation Code

A.1 Actuator.cpp

Приложение B: Controller Code

B.1 Control.cpp

B.2 Def.h

B.3 Hardware.cpp

B.4 Predict.dat

B.5 Gain.dat

Приложение C: Robot Hardware

C.1 Strain Gage Amplifier

C.2 Wiring

C.3 Mechanical Drawings

Литература

---

Список таблиц

Таблица 4.1 : Transducer sensitivity and error

Таблица 6.1 : Joint range of motion

Таблица 6.2 : Time parameters for walking motion

Таблица 6.3 : Walking motion control gains

Таблица 6.4 : Passivity and average duty cycles of each valve

---

Список иллюстраций

Иллюстрация 1.1 : Dimensionless force-length properties of actuators and biological muscles

Иллюстрация 2.1 : Photograph of inflated and uninflated actuators

Иллюстрация 2.2 : Actuator dimensions

Иллюстрация 2.3 : Geometric schematic of actuators

Иллюстрация 2.4 : Mesh geometry

Иллюстрация 2.5 : Revised actuator geometry schematic

Иллюстрация 2.6 : Pressure, Length, Force, Stiffness Relationship for a BPA

Иллюстрация 2.7 : Static model verification schematic

Иллюстрация 2.8 : Plot of Force vs. Length for a BPA with constant internal mass of air

Иллюстрация 2.9 : Pressure Increase vs. Length – constant mass system

Иллюстрация 2.10 : Plot of Force vs. Length – constant air mass – Exp. vs. Theor. results

Иллюстрация 2.11 : Plot of Effectiveness vs. Pressure

Иллюстрация 2.12 : Plot of Force vs. Length – Exp. vs. Theor. Results – (effectiveness)

Иллюстрация 2.13 : Plot of Force vs. Length – Constant Pressure System

Иллюстрация 2.14 : Dynamic model schematic

Иллюстрация 3.1 : Simulation overview schematic

Иллюстрация 3.2 : Detailed simulation schematic

Иллюстрация 3.3 : Actuator equation of motion schematic

Иллюстрация 3.4 : Flow curve for Matrix 758 3-way valve

Иллюстрация 3.5 : Dynamic model verification schematic

Иллюстрация 3.6 : Length vs. Time – 60 psi constant mass – 6 lb load - measured

Иллюстрация 3.7 : Length vs. Time – 60 psi constant mass – 6 lb load - simulation

Иллюстрация 3.8 : Length vs. Time – 80 psi constant mass – 11 lb load - measured

Иллюстрация 3.9 : Length vs. Time – 80 psi constant mass – 11 lb load - simulation

Иллюстрация 3.10 : Length vs. Time – 60 psi constant mass – 11 lb load - measured

Иллюстрация 3.11 : Length vs. Time – 60 psi constant mass – 11 lb load - simulation

Иллюстрация 3.12 : PWM valve model verification schematic

Иллюстрация 3.13 : Commanded vs. Actual duty cycles – Matrix 758 3-way valve

Иллюстрация 3.14 : Length vs. Time – 100 psi - 25 Hz, 50% PWM – 5 lb load - measured

Иллюстрация 3.15 : Length vs. Time – 100 psi - 25 Hz, 50% PWM – 5 lb load - simulation

Иллюстрация 3.16 : Length vs. Time – 100 psi - 25 Hz, 50% PWM – 1 lb load - measured

Иллюстрация 3.17 : Length vs. Time – 100 psi - 25 Hz, 50% PWM – 1 lb load - simulation

Иллюстрация 3.18 : Length vs. Time – 100 psi - 25 Hz, 50% PWM – 15 lb load - measured

Иллюстрация 3.19 : Length vs. Time – 100 psi - 25 Hz, 50% PWM – 15 lb load - simulation

Иллюстрация 3.20 : Length vs. Time – 100 psi - 50 Hz, 50% PWM – 1 lb load - measured

Иллюстрация 3.21 : Length vs. Time – 100 psi - 50 Hz, 50% PWM – 1 lb load - simulation

Иллюстрация 3.22 : Length vs. Time – 100 psi - 50 Hz, 50% PWM – 5 lb load - measured

Иллюстрация 3.23 : Length vs. Time – 100 psi - 50 Hz, 50% PWM – 5 lb load - simulation

Иллюстрация 4.1 : Robot hardware schematic

Иллюстрация 4.2 : Photograph of robot

Иллюстрация 4.3 : CAD model and photograph of robot leg

Иллюстрация 4.4 : CAD model and photograph of hip translational joint

Иллюстрация 4.5 : CAD model and photograph of hip rotational joint

Иллюстрация 4.6 : Schematic of inlet valve

Иллюстрация 4.7 : Schematic of exhaust valve

Иллюстрация 4.8 : Current vs. time for inlet valve

Иллюстрация 4.9 : Current vs. time for exhaust valve

Иллюстрация 4.10 : Force sensor classic analysis schematic

Иллюстрация 4.11 : Force sensor FEA results

Иллюстрация 4.12 : Photograph of force sensors

Иллюстрация 4.13 : Photograph of strain gage amplifier

Иллюстрация 4.14 : Transducer calibration plot

Иллюстрация 4.15 : Photograph of completed force measurement system

Иллюстрация 4.16 : Photograph of completed angle measurement systems

Иллюстрация 5.1 : Labeled schematic of a joint

Иллюстрация 5.2 : Block diagram of the control algorithm

Иллюстрация 5.3 : ISR schematic

Иллюстрация 5.4 : Inverse kinematics schematic

Иллюстрация 6.1 : Desired foot positions for walking motion

Иллюстрация 6.2 : Sequential video frames of the leg during walking motion

Иллюстрация 6.3 : Desired and actual x-y foot paths: Test 1

Иллюстрация 6.4 : Desired and actual joint angles vs. time: Test 1

Иллюстрация 6.5 : Actuator force vs. time

Иллюстрация 6.6 : Desired and actual joint stiffness vs. time

Иллюстрация 6.7 : Joint torque vs. time

Иллюстрация 6.8 : Ground reaction forces during walking vs. time

Иллюстрация 6.9 : Trolley motion vs. time

Иллюстрация 6.10 : Hip joint valve duty cycles vs. time

Иллюстрация 6.11 : Knee joint valve duty cycles vs. time

Иллюстрация 6.12 : Desired and actual joint angles vs. time: Test 2

Иллюстрация 6.13 : Desired and actual x-y foot paths: Kicking motion

Иллюстрация 6.14 : Desired and actual joint angles vs. time: Kicking motion

Иллюстрация 6.15 : Calculated angular velocity vs. time

Иллюстрация 6.14 : Desired and actual joint angles vs. time: Derivative control

Иллюстрация 7.1 : Desired and actual x-y foot paths: Angle feedback only

---

Глава1

Acknowledgements

I would like to thank my wife Joni for being supportive of this crazy idea to go get a masters degree. Without her support, I never would have been able to do it. I would also like to thank my wonderful new son Boyd for helping me put priorities into perspective.

I want to thank Roger Quinn for providing an excellent work environment. I have thoroughly enjoyed my experience working in the Biorobotics Lab. I also appreciate his patient and supportive leadership style.

Gabe Nelson has given me invaluable counsel in robotics as well as life. In this thesis, you will see evidence of all the work he has done before me to make my work possible. For this, I would like to thank him.

I want to thank Rich Bachmann, Matt Birch, Jim Berilla, and Yuan Dao Zhang for using their time and talents to help me build the robot.

I would also like to express my gratitude to the other guys in the lab for being sounding boards for my half-baked ideas and providing a good political discussion at the drop of a hat.

I want to express my appreciation to my parents for raising me and giving me direction. I was taught many of the fundamentals of engineering while tinkering in the garage with my dad.

I want to thank Dr. Joe Mansour and Dr. Steve Phillips for serving on my thesis committee.

The work was supported by the Office of Naval Research under grant number

N0014-99-1-0378, and DARPA under contract number DAAN02-98-C-4027.

Finally, I want to thank God for life, and all these wonderful creatures that we study. Even with all our great technology, our understanding of the world is so limited. As an engineer, His infinite wisdom and designs humble me. Without Him, we are nothing

---

Design and Control of a Robotic Leg With Braided Pneumatic Actuators Abstract by ROBB WILLIAM COLBRUNN A Braided Pneumatic Actuator is a device developed in the 1950’s by J.L.

McKibben. In recent years, robotics engineers have begun to rediscover these fascinating devices, and use them as actuators for their robots. These actuators exhibit non-linear force-length properties similar to skeletal muscle, and have a very high strength-to-weight ratio. In this thesis, emphasis was placed on understanding the actuator properties so that this knowledge could be used in simulation and control of legged robots. Static and dynamic mathematical models were developed for the actuators, and verified through testing and simulation. A four-degree of freedom robotic leg was designed, constructed, and controlled. The leg provided stable, sensible forward walking for the robot, and was capable of operating 94% passively. Though these actuators have a few limitations, their muscle-like properties including high strength-to-weight ratio, passive characteristics, and self-limiting force properties make them ideal for legged robots.

---

Chapter I: Introduction 1.1 Background

Much ink has been used to explain the benefits of legged locomotion over wheeled locomotion for robots. As with most things, there is always a trade off. A wheeled robot might go faster and be easier to build, but tell it to climb a flight of stairs and its shortcomings become apparent quickly. Upon observation of all the creatures around us, we see the legged design as the locomotion method of choice. The purpose of this thesis, however, is not to show why it should be done, but how it can be done.

Legged locomotion is the beginning of the biologically inspired approach to robotics. However, putting legs on a frame is not enough to have a walking robot. Arguably, the most important part of any motion device is the mechanism that does the moving: the actuators. In the past, CWRU walking robots have used DC motors to generate leg movement. With the move toward physical autonomy (operation without an umbilical), the poor strength to weight ratio of DC motors became far too prohibitive. The next generation of robot used pneumatic cylinders to increase this ratio. Air cylinders’ strength to weight ratio is much better than electric motors, but there is still room for improvement. The total weight of Robot III was 29.5 lbs. with 76% of that being actuator and valve weight (Bachmann, 2000). A lighter weight actuator that was just as strong would increase the robot’s payload capacity. The biologically inspired approach led the group to look at the prime movers that are designed into natural walking “machines”: muscles. Using real muscle tissue is technologically not practical at this time (Shimoyama, 1997), so an artificial muscle is necessary. This actuator is known as a Braided Pneumatic Actuator or a McKibben Artificial Muscle. These actuators can have a power-to-weight ratio as high as 25 times greater than DC motors (Shadow, 2000). These actuators also have experimentally been shown to have force-length properties similar to skeletal muscle (Klute et al.,1999).

Figure 1-1: The dimensionless relationship between force and length under isometric conditions at maximal activation for various animals as well as a McKibben actuator pressurized to 5 bar (Klute et al.,1999).

This muscle-like stiffness and their structural flexibility present real advantages over air cylinders. In this thesis, these braided pneumatic actuators will be referred to as “McKibbens” or BPA.

1.2 Purpose and Overview

The focus of this project involves the investigation of using McKibbens as the prime mover in a walking robot. This work is divided into six sections. Each chapter builds upon the other to explain the process and methods of using BPA to make a robotic leg walk. Chapter two explains all the properties of the BPA as well as the mathematical model formulated for simulation and control. Chapter three describes the simple spring- mass-damper simulation that was written to dynamically validate the model. This simulation included a model of the actuator, and a model of a solenoid valve for controlling the mass of air in the actuator. The valve states were the only inputs, and the simulation calculated the remaining system properties. The next chapter describes all the hardware involved in making the robot walk. A system overview, the valves, the sensors, and the leg design are all presented. Chapter five delves into the control program and how the sensor feedback is turned into valve commands. Then, chapter six presents the results or quantification of the walking. The last chapter discusses the limits and shortcomings of using McKibbens. Finally, possible solutions and improvements are addressed as topics of future work.