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)?
 ..
