Professional Documents
Culture Documents
Yasser F. O. Mohammad
Which is harder?
To decide which
to move
To actually
piece
move it
Overview
Introduction to the field of
Basic
Problems in Robotics
Robotic Architectures
Imitation in Robotics
ISR
What is a robot?
The word robot came from Karel Capeks
2013 BAXTER
2015 Cat
2004
Leonardo
2014 Nextage
2004 Robonaut
2015 ATLAS
2006 NAO
ASIMO
2002
Roomba
1961 Unimate
1978 SCARA
Research Fields
Machine
Learning & AI
HumanComputer
Interaction
Automatic
Control
Psychology
(Developmenta
l,
Experimental,...
)
Neuroscience
Social Science
Mechanical
Design &
Robotics
Related Fields
Subfields
IJSR: Social Roboticsis the study of robots that are able to interact and
communicate among themselves, with humans, and with the environment,
within the social and cultural structure attached to its role.
Compliance, safety and compatibility in the design of social robots ''living'' with humans
Edutainment robotics
Industr
ial
Robot
Static
and
fully
predict
able
Task
s
Fixed
and
repetiti
ve
Variable
and
evolving
In
Examples:
Given a RR robot with angles 1 and 2 find the location and
orientation of the end effector
Given
a RRPRRPRPPPPRPRRRRPPPRPRPPPRPRPRPPPRRPRP
manipulator, find the location and orientation of the end effector
In
general:
We use Denavit-Hartenberg convention for frame
assignment
We use transformations between frames to get to the tool
Examples:
How to make the end effector touch a box?
How to make the end effector reach a point?
Elbow down
Elbow Up
Example
Reflection
Deliberation
GOFAI
Shakey
70s~80s
Very Slow
Imitation
Reactivity
ML
Baxter
2000~
challenging
Reinforcement
Learning
90s~
Not scalable
Sense
data
Strips
Action
s
Directive
s
Shakey
by SRI for DARPA 1967
70
Example Strips
Difference
Table
Summary
Example: Subsumption
Hybrid Architectures
Deliberative
=
Hybrid
Reactive
Example: AuRA
Example: EICA
Execution Time
Activation Level
Behavioral Influence
Intentions
Mutation
Eliting
Tournament
Cross Over
Mutation
Learn a model
Reproduce the task using the model
Data
Mining
Vision/
Robotics
ML
Social
sciences
Free Context
Fluid
Imitation
Few demonstrations
Known
Primitives
Standard ML
NN/SVM/HMM
Planned
Kinesthetic Teaching
Motif Discovery
Correspondence
Problem
Unplanned
Learning from Observation
Taxonomy of LfD
Learning from
Demonstration
Learning primitives
DMP
GMM/GMR
Inverse RL
Skill Trees
Fluid imitation
Bayesian IL
GMM/GMR
Skill learning in RL
IOC (1997)
User to learn ball balancing from a single demonstration
GMM/GMR (2005)
GMM/GMR example
Example of GMM/GMR
Recent Advances
Diritchilet
2013
Variational
2015
Operational
2012
Serialization
2013
Cyclic motion
2012
GMR
2013
GMM/GMR
Multi Frames
2009
SAXGMM
2014
Piecewise
linear (2015)
QGMM
2012
PHMM
2010
DMP
Limear/Cyclic
2013
SAXImitate
SAXDMP
(2015)
Fluid Imitation
Current State-Of-the-Art
Focus on
MOTION/OBJECTS
Focus on
INTERACTION/PEOPLE
Collaboration
Focus on HOW
Fluid Imitation
Example
Learned Behaviors
Saliency (scenario)
Relevance (scenario)
Experimental Setup
Results
Gestures Discovery
rate (5/7)
Action Discovery rate (7/7)
Action-Gesture Association Accuracy: 71.4%
Learning to Write
11 DoF redundant serial manipulator
ODE physics simulation
User writes 1452 words article in Arabic
on a touch
screen device
X-Y coordinates are recorded as input to FIE
Inverse Kinematics was learned using a babbling stage.
Results
References
Intelligent Robotics
Reference
Online Courses