2.12 Final LectureHumanoids and Biologically-Inspired Robots
Where are we going in 10 ~ 25 years?2.12 Final Lecture
H. Harry Asadad’Arbeloff Laboratory for Information Systems and Technology
Department of Mechanical Engineering M.I.T.
Humanoids: Can we make our own copy?
Four photos of humanoid robots removed for copyright reasons.
ButteryActuatorsTransmission
Biped locomotion
Anatomy, Physiology, Biomechanics
Kinematics, Dynamics, Control
BehaviorSoftware architectureVisionSpeech recognition
Skills
Biped Locomotion: How does a human walk?
Biomechanics Model Human measurement Kinematics, dynamics, and control
Photo removed for Photo removed for
copyright reasons. copyright reasons.
Fundamentals of biped locomotion
Torso
Gravity
Inertial Force Acceleration
Reaction Force from the Floor
Zero Moment Point (ZMP)
The body may tip over, if the floor reaction varies.
Gravity
Inertial Force Acceleration
Actual Reaction Force
Misalignment Æ Tip Over
Bump, object, or roughness
from the Floor
Zero Moment Point(ZMP)
1. Floor Reaction Control
Zero Moment Point
(ZMP) Accommodate the floor reaction force by
Gravity
Inertial Force Acceleration
distributing the force between the heel and the toe, so that the resultant force may pass
through the ZMP
2. Target ZMP Control
Accelerate the upper torso to increase the inertial force so that the resultant force may pass through the ZMP.
Gravity
Inertial Force Increased Acceleration
New ZMP
Original Zero Moment Point(ZMP)
3. Stride Control
Pre-planned target torso position
As a result of the ZMP control, the target position of the upper torso is shifted towards the direction of acceleration.
ZMP
The stride is accommodated to keep up with the torso speed
Stereo Vision
3-axis accelerometers+ Gyro
6 DOF
Joint angle sensors
Photo removed for copyright reasons: side view of Honda
ASIMO robot. http://world.honda.com/
ASIMO/
Battery Controller RF Link
Compliance Control
6-axis force/torque sensors
Amputee with force feedback. MIT Leg Lab/Media Lab, Hugh Herr
Photo removed for copyright reasons.
Hami Kazerooni’s Robotic ExoskeletonUC Berkeley, Human Engineering and Robotics Lab
Photos removed for copyright reasons.
Source: UC Berkeley Robotics and Human Engineering Laboratory.http://bleex.me.berkeley.edu/bleex.htm
5 body-lengths per second
Stanford Cockroach Robot
Figures removed for copyright reasons.
Mark CutkoskyBiomimetic Design and Fabrication of a Hexapedal Running Robot
Stanford University, Center for Design Research
J. Clark, J. Cham, S. Bailey, E. Froehlich, P. Nahata, R. Full (UC Berkeley, Biology), M. Cutkosky, Proc. 2001 IEEE ICRA
Figures removed for copyright reasons.
Biologically-Inspired Robots
(Courtesy of MIT. Used with permission.)
Robo-Pike at MITPike can accelerate at a rate of 8 ~ 12 G’s.
(http://web.mit.edu/towtank/www/Pike/pike.html)
Snake Robots by Shigeo Hirose
(Courtesy of Prof. Shigeo Hirose. Used with permission.)
Biologically-Inspired Robots
Photo courtesy of Los Alamos National Laboratory. (Courtesy of MIT. Used with permission.)
Dinosaur Robot, Troody Spyder 1.0 by Mark Tilden
Still long way to go.
What are missing?Brain and Muscle
Photo removed for copyright reasons: promotional image for film I Robot.
Actuator Research and Development:Very slow progress
Progress
Computer
Sensor
Actuator
Year
Sensors, everywhere …..
Actuators, everywhere …
Sensor
Actuator
1995 2005
MITTom Swager, Solder Nanotech Lab & Chemistry
Ian Hunter, Bioinstrumentation Lab, Mechanical Engineering
Diagrams removed for copyright reasons.
Initial results on a new conducting polymer actuator show strains of 6% against an applied load of 1 MPa at a strain rate of 1% per second. The material is Poly EDOT and it is actuated electrochemically in the ionic liquid BMIMBF4. The results will be presented in two weeks at the SPIE 11th Annual International Symposium on Smart Structures and Materials.
Material Based Artificial Muscle Actuators
ShapeShape ElectroactiveElectroactive CarbonCarbon MemoryMemory PolymerPolymer NanotubeNanotubeAlloy ActuatorsAlloy Actuators ActuatorsActuators ActuatorActuator
HN
NH HN
NH HN
Temperature
Strain Photos removed for copyright reasons.
Courtesy of NASA JPL.
Actuation Mechanism: Uses Shrinkage or Expansion of material to create displacement
There are several kinds of Material based artificial muscle actuators, fromshape memory alloys to eap, and most recently carbon nanotube actuators. These actautors share a conceptually common actuation mechanism, which is shrinkageor expansion of material to create a strain.
Actuator Characteristics
Energy Density Stress x Strain
Cross-sectional Area: A
Force: F Strain ε
These graphs show the comparison of material characteris of these actautors,against natural muscles, energy densiy, specific power.
Actuators Compared with Material Characteristics of Muscle
And this table shows you more of the material characteristics of these actuators. And these characteristics dictatesthe implementation of these actuators into real systems. The point I would like to make here, is that lots ofresearchers are interested in the material characteristics of these actuators and tries to exploit these charactericswhen designing a system. But how about the architectural characteristics of muscles? Is there anything interesting or useful in the architectural characteristics of muscles?
Robotics: Systems
Fill the technology gap We cannot wait!
Materials Technology
Shape Memory AlloyShape Memory AlloyThe largest stress and energy density among all the actuator materials
> 200 MPa> 4x107 J/m3 Hysteresis
Strain
cooling
Temperature
Joule heating and forced air cooling
Shape Memory AlloyShape Memory Alloy
The hysteresis curve shifts depending on the stress applied
Hysteresis cooling
Strain
Joule heating and forced air cooling
Stress
Temperature
Shape Memory AlloyShape Memory Alloy
The hysteresis curve shifts depending on the stress applied
Strain
Hysteresis
Stress
Joule heating and forced air cooling
cooling
Temperature
The Traditional Approach is “BULK” FEEDBACK.
Control Actuator Material
Since the process is highly distributed and nonlinear, bulk feedback does not work well.
The Traditional Approach to these systems is BULK FEEDBACK.
Control Actuator Material
Since the process is highly distributed and nonlinear, bulk feedback does not work well.
New Approach:
A c t u a t o r M a t e r i a l
Segmented Binary Control
Local Control
Segment-by-segment simple local controls (binary, finite state controls)
Muscles have Segmented Architecture
Whole Muscles Long strap muscle
Muscle fibers Unit of independent Innervation
Myofibrils Muscle thread
Functional unit ofSarcomeres contraction
Myosin, actin Components of sarcomeres
Figures by MIT OCW.
One interesting point is that muscles have segmented architecture. This is a picture from basic human anatomy book, and as you can see, muscles are segmented in several hierachy. It starts with myosin, and actin, which are components, and sarcomeres are put in series to form myofibrils, which is a muscle thread.
Regardless the size of animal, the building block is the same!
Figure by MIT OCW.
Segmented Binary Control
• Divide the whole into a multitude of smaller segments controlled separately.
• Overall strain is the sum of the individual strains for each segment
Temperature
Strain
Segmented
ONOFF
Binary State Controls
• Each segment may take either hot state or cold state
• Wash out all material hysteresis and nonlinearities
StressNormalized Displacement
0
0.5
1.0
C D
A
B
Increase
TM Temperature TA
Phase transition diagram of SMA and selection of threshold temperature TA and TM
The hypothesis.Based on expected stress states in the design, hot and cold temperatures can be chosen.
Implementation of Segmented Binary ControlSMA Use of Thermoelectric devices (Peltier Effect) for
selective, local heating and coolingselective, local heating and cooling
Tension
SMA Wire Thermoelectric Devices sandwiching the SMA wire
Use of Thermoelectric devices (Peltier Effect) for selective, local heating and coolingselective, local heating and cooling
Heating
−
+ Cooling
−
+
Schematic diagram removed for copyright reasons.Source: http://www.tellurex.com.
Dis
plac
emen
t (m
m)
Segmented Binary Control:SMA works like a stepping motorstepping motor
9
8
7
6
5
4
3
2
1
0 0 1 2 3 4 5
Number of Elements "On"
Load = 0 Load = 1 N Load = 2N Load = 3N Load = 4.5N
X =η n N
η ≈ 6.1 mm
Con
trol
Con
trol
Con
trol
Con
trol
Con
trol
Con
trol
Con
trol
Con
trol
Drawback: Too many controls
16 units
Con
trol
Con
trol
Solution: Grouping
ControlControl Control
1 2 units 4 units 8 units
Minimum segmentation of single axis
1248
D t S r
S e
3 3
2 2
1 1
0 0 2222 aaaa +++
isplacemenenso
MA Wir
Can we make more saving?Particularly for multi axis actuators
S1 S2 S3 SN
A1
S1 S2 S3 SN
A2
M × N segments
S1 S2 S3 SN
AM
Too many segments Couple the segmentsfor multi-axis case
Disadvantage of segmented binary control
Two-dimensional segmentation of multi-axis SMA actuator system on a Peltier pellet bed
CoupledFixture SMA wire bundle
Axis 1
Axis M
segments
Independent Dependent segmentssegments
Array Actuators
Muscles are Coupled
Deep anterior musclesof the arm
Several muscles for single motion Several Muscles with Different Functions Work together to create single DOF (Fine motion and Gross Motion)
Muscle for Gross Motions
Muscles for Fine Motions
Single muscle for several motions Some muscles are connected to multiple bones
Flexor digitorum profundus: connected to four fingers
Flexor digitorum superficialis
Figure by MIT OCW. This is a diagram of deep anterior muscle of the arm. Please note that several muscles are used for activating single DOF, meaning several muscles are attahed to the same bone, and sometimes single muscle is connected to multiple bones. For example this gross motion
Muscles for Fine Motions
Gross Motions
profundus
Muscles for
F lexor digitorum
Figure by MIT OCW.
Coupled SegmentsIndependent Segments
Embodiment of the muscle-tendon-bone coupled architecture
Architectural Characteristics
Conventional Architecture
Current Muscle-Like actuators use this architecture
1. Actuators as a whole 2. One actuator per each degree of freedom
Muscle-like Architecture
1. Segmented architecture2. Coupled architecture
Multiple actuators for each degree of freedom and some actuators drive several d.o.f.
So in summary so far, the architectural characteristics of actuators using conventional approach is that each actuator is treated as a whole, and there is only actuator per each degree of freedom. Where as muscle-like approach would be to have a segmented architecture, and coupled architecure. And systems developed with artificial muscle actuators don’t necessarily have a muscle-like architecture.
Implementation using Thermoelectric Devices
Top View of Actuator Array built with SMA sandwiched between Thermoelectric Modules
Postures of the human hand
PINCH1 PINCH4
POINT
For daily manipulative tasks we use only a limited number of hand postures.
OPEN SURVEY ENVELOPE1 ENVELOPE2
BALL GRIP1 BALL GRIP2 FIST GRIP1 FIST GRIP2
PINCH2 PINCH3
WRITE
Photos removed for copyright reasons.
Optimal Design: Minimum Segmentation
Two-dimensional segmentation of multi-axis SMA actuator system for Robotic Hand
Segmentation Designed to Minimize the Number of Segments Based on Trajectory of Each Actuator Total of 21 Segments → reduction from 50 segments
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Each posture can be represented by 21 bits of 0’s and 1’s
1.OPEN
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
2.SURVEY
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
3.ENVELOPE1
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
4.ENVELOPE2
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
5.BALL GRIP1
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
6.BALL GRIP2
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
7.FIST GRIP1
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
8.FIST GRIP2
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
9.PINCH1
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
10.PINCH2
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
11.PINCH3
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
12.PINCH4
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
13.POINT
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
14.WRITE
Axis 3
Axis 2
Axis 4
Axis 7
Axis 6
Axis 1
Axis 9
Axis 8
Axis 10
Axis 5
Photo removed for copyright reasons.
10 Axes of actuator array controlled by 12 ON-OFF controllers
Peltier Effect Thermoelectric Devices Programmable Array
Jumper Cable Terminals
Five-Fingered HandWith 10 DOF Actuators
Power Grip 10 Axis SMA Wires
Temperature Sensors
Coupled Motions
Coupled Motions
Cellular Actuators
• Think small! Control of small building blocks
• Think big! A vast number of DOF streamlined
• Think simple! Finite state machines: Software
• Think as a system! A system exhibits something special
Cognition: Another grand challenge
Photo of MIT's Cog robot removed for copyright reasons.See http://www.ai.mit.edu/projects/humanoid-robotics-group/cog/cog.html.
Home Robotics in Japan
Photos removed for copyright reasons.See http://www.menzelphoto.com/gallery/big/robo2.htm.
Home Robotics
NEC Personal Assistant Robot R100
Photo removed for copyright reasons.
Photo removed for copyright reasons.
Your mom said, “…”
Pet robot, “Tama”, can comfort and heal the elderly and the disabled
UbiquitousHome
Robotics
H. Asada, MIT Exoskeleton Rehabilitation Suit
FingernailSensors for Human-Machine Interface Ring Sensor
Repositioning Active Bed Sheet
Wearable HealthMonitoring
Health Chair
Monitor pilot light in furnace Monitor fire
alarms
Monitor heating and
cooling
Monitor windows &
doors
Pet Monitoring
Home Appliance Maintenance
Home Security
Robot Market
• Home robotics – Vacuum cleaner robots – Entertainment – Hobby/education
• Elderly care • Security
Santa’s Home Delivery Robot2.12 Introduction to Robotics
Final Project
Final Project Demonstration• 2:30 Meet in the 2.12 Lab
Final check/preparation • 2:50 Move your machine to the pit in
front of the graduate machine shop • 3:00 Welcome and introduction, Chorus• 3:10 Task 1 and Task 2 Demo
Have the robot run from the North Station to the Village • 3:25 Task 3 and Task 4 Demo within the Village• 3:40 Task 5 and Task 6 (if presentable) • 3:50 Award presentation
» Best Group Award: 2.12 Robot Hall-of-Fame » Rudolph Prize (MVP)
• 4:00 Refreshments
2.12 Lab Pappalardo Lab Entrance
North Pole StationElevator
Robot Pits
Doll House Village
Polar Express Way d’Arbeloff Lab Entrance
2.12 Lab Doll House Village
No. 9 Magnet First Doll House
Chimney (Second Doll House)
Third Doll House Polar Express Way