Transcript Document
Robotic Self-Perception and
Body Scheme Learning
Jürgen Sturm
Christian Plagemann
Wolfram Burgard
University of Freiburg
Germany
With annotated questions
SA-1
Outline
Presentation of current research
Clarification of Concepts, e.g.,
- Mirror Neurons
- Body Schema
Planned journal articles
Possible support of our theory
from a psychological point of view?
Future research / experiments
Psychological evidence
Co-experiments human/robot
General Brainstorming
Motivation
Existing robot models are typically
specified (geometrically) in advance
calibrated manually
Motivation
Problems with fixed robot models:
Wear-and-tear
wheel diameter, air pressure
Recovery from failure
malfunctioning actuators
Tool use
extending the model
Unknown model
re-configurable robots
Motivation
Problems with fixed robot models:
Wear-and-tear
wheel diameter, air pressure
Recovery from failure
malfunctioning actuators
Tool use
extending the model
Unknown model
re-configurable robots
Similar problems in humans/animals?
Motivation
Problems with fixed robot models:
Wear-and-tear
wheel diameter, air pressure
Recovery from failure
malfunctioning actuators
injured body parts
Tool use
extending the model
growth, aging
writing
Unknown model
re-configurable robots
riding a bike
Similar problems in humans/animals?
Related Work
Neuro-physiology
Mirror neurons
Body Schemes
Clarification of concepts
Better references?
Good primer?
[Rizzolatti et al., 1996]
[Maravita and Iriki, 2004]
Robotics
Self-calibration [Roy and Thrun, 1999]
Cross-modal maps [Yoshikawa et al., 2004]
Structure learning [Dearden and Demiris, 2005]
Where else? E.g.,
- Self-configuring software
- Language acquisition
Problem motivation
Fixed-model approaches fail when
parameters change over time
geometric model is not available
Our Contribution
Bootstrapping of the body scheme and
Life-long adaptation using visual
self-observation
Problem Description
Think
Bootstrap, monitor, and maintain
internal representation of body
Self-observation
Sense
6D Poses
Motor babbling
Act
Joint angles
Problem Formulation
Visual self-perception of n body parts:
X 1 ; : : : ; X n 2 R4£ 4
Actuators (m action signals):
a1 ; : : : ; am 2 R
Learn the mapping
Which brain area
does this mapping?
p (X 1 ; : : : ; X n ja1 ; : : : ; am )
Body pose
Configuration
Existing Methods
Analytic model + parameter estimation
Requires prior knowledge
Function approximation
Nearest neighbor
Neural networks
High-dimensional learning problem
Requires large training sets
Body Scheme Factorization
Idea: Factorize the model
Local models
similar to
mirror neurons?
We represent the kinematic chain as a
Bayesian network
Bootstrapping
Learning the model from scratch consists of
two steps:
1. Learning the local models (conditional
density functions)
Mirror neurons?
2. Finding the network/body structure
Synaptic pathways?
Learning the Local Models
Using Gaussian process regression
Learn 1D 6D transformation function
p(¢ 12 j a1 ) = p(X ¡ 1 X 2 j a1 )
1
for each (action, marker, marker) triple
Finding the Network Structure
Select the most likely network topology
Corresponding to the minimum spanning
tree
Maximizing the data likelihood
p(M jD)
Model Selection
Model Selection
7-DOF example
Fully connected BN
Model Selection
More natural,
incremental algorithm?
E.g., simulated
7-DOF
synapticexample
growth..
Fully connected BN
Selected minimal
spanning tree
Forward Kinematics
Purpose:
prediction of end-effector pose in a given
configuration
Approach:
integrate over the kinematic
chain in the Bayesian network
by concatenating Gaussians
approximate the result
p (Xefficiently
jX ; a ; :by
: : ;one
a )Gaussian
=
Z
n
m
Z1 1
: : : pM pM : : : dX 2 ; : : : ; dX n ¡
1
2
1
Inverse Kinematics
Purpose: Generate motor commands for
reaching a given target pose
Approach: Estimate Jacobian of endeffector using forward kinematics prediction
·
r X n (a)
=
@X n (a)
@X n (a)
;:::;
@a1
@am
Use standard IK techniques
Jacobian pseudo-inverse
¸
Experiments
Evaluation: Forward Kinematics
Fast convergence (approx. 10-20 iterations)
High accuracy (higher than direct perception)
Evaluation: Inverse Kinematics
Accurate control using bootstrapped body
scheme
Life-long Adaptation
Robot’s physical properties Physiology?
Anatomy?
will change over time
Predictive accuracy of body scheme
needs to be monitored continuously
Body Schema Plasticity in humans/animals
Localize mismatches in the Bayesian
network
Re-learn parts of the network
Life-long Adaptation
Similar problem?
Recovery after
lesions to the
brain?
Initial
Error is detected and is localized
Robot re-learns some local models
Life-long Adaptation
Evaluation
Recovery time
plot for a human
after body
“deformation”?
Quick localization of error
Robust recovery
Summary
Novel approach learning body schemes
from scratch using visual self-perception
Model learning using Gaussian process
regression
Model selection using data likelihood as
criterion
Efficient adaptation to changes in robot
geometry
Accurate prediction and control
Future Work
Active self-exploration, optimal control,
POMDPs
Marker-less self-perception
Moving robot
Tool use
Future work: Tool Use
Using tools requires dynamic extensions of
the body scheme
Future research / experiments
Tool use
Writing with a pen
Approach:
•
•
•
•
Find Silhouette of Pen
Detect tool-tip
Assume rigid tool
Learn geometric transformation
Demonstration:
• Write/paint on whiteboard with pens of different size
and shape
Student projects (1)
Tutoring Evasion Maneuvers using Tactile
Sensors (Frederic Dijoux)
Student projects (2)
Model-Free Control for Robotic Manipulators
using Nearest-Neighbor methods (Hannes
Schulz and Lionel Ott)
Student projects (3a)
Dynamically adding repellant end-effectors
(Clemens Eppner)
Student projects (3b)
Programming by Demonstration (Clemens
Eppner)
Student projects (3c)
Programming by Demonstration (Clemens
Eppner)
Student projects (3d)
Programming by Demonstration (Clemens
Eppner)
Student projects (4a)
Object Recognition using Tactile Sensors
(Alexander Schneider)
Student projects (4b)
Object Recognition using Tactile
Sensors (Alexander Schneider)
Student projects (5a)
Grasping objects using Visual Servoing
(Nikolas Engelhard)
(Video courtesy of TU Dortmund)
Student projects (5b)
Grasping objects using Visual
Servoing (Nikolas Engelhard)
(Video courtesy of TU Dortmund)
Planned journal article
Special Issue “Journal of Physiology”
Neuro-Robotics Symposium – Sensorimotor
Control, July 2008, Freiburg
Two reviewers, one from neuro-biology, one
from engineering, Deadline: 22.10.2008
Article Content:
Similar to this presentation
Stronger focus on mirror neurons and body
schemas in humans/animals
Support from psychological point of view?
Possible journal article
Special Issue “Autonomous Mobile
Manipulation”
Journal “Autonomous Robots”
Deadline: 15.12.2008
Article Content (if at all):
Focus on Model selection?
..
Brainstorming
Psychological input
Co-experiments human/robot
Joint (student) project(s)?
..