Robotic Self-Perception and Body Scheme Learning

1
Robotic Self-Perception and Body Scheme Learning

• Jürgen Sturm
• Christian Plagemann
• Wolfram Burgard
• University of Freiburg
• Germany

2
Motivation

• Existing robot models are typically
• specified (geometrically) in advance
• calibrated manually

3
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

4
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?

5
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?

• growth, aging
• injured body parts
• writing
• riding a bike

6
Related Work

• Neuro-physiology
• Mirror neurons Rizzolatti et al., 1996
• Body Schemes Maravita and Iriki, 2004
• Robotics
• Self-calibration Roy and Thrun, 1999
• Cross-modal maps Yoshikawa et al., 2004
• Structure learning Dearden and Demiris, 2005

7
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

8
Problem Description
Think Bootstrap, monitor, and maintain internal
representation of body
Act Joint angles
Sense 6D Poses
9
Problem Formulation

• Visual self-perception of n body parts
• Actuators (m action signals)
• Learn the mapping

Configuration
Body pose
10
Existing Methods

• Analytic model parameter estimation
• Function approximation
• Nearest neighbor
• Neural networks

Requires prior knowledge
High-dimensional learning problem
Requires large training sets
11
Body Scheme Factorization

• Idea Factorize the model

We represent the kinematic chain as a Bayesian
network
12
Bootstrapping

• Learning the model from scratch consists of two
  steps
• Learning the local models (conditionaldensity
  functions)
• Finding the network/body structure

13
Learning the Local Models

• Using Gaussian process regression
• Learn 1D ? 6D transformation functionfor each
  (action, marker, marker) triple

14
Finding the Network Structure

• Select the most likely network topology
• Corresponding to the minimum spanning tree
• Maximizing the data likelihood

15
Model Selection
16
Model Selection

• 7-DOF example
• Fully connected BN

17
Model Selection
7-DOF example Fully connected BN Selected
minimal spanning tree
18
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
• efficiently by one Gaussian

19
Inverse Kinematics

• Purpose Generate motor commands for reaching a
  given target pose
• Approach Estimate Jacobian of end-effector using
  forward kinematics prediction
• Use standard IK techniques
• Jacobian pseudo-inverse

20
Experiments
21
Evaluation Forward Kinematics

• Fast convergence (approx. 10-20 iterations)
• High accuracy (higher than direct perception)

22
Evaluation Inverse Kinematics

• Accurate control using bootstrapped body scheme

23
Life-long Adaptation

• Robots physical properties will change over time
• Predictive accuracy of body scheme needs to be
  monitored continuously
• Localize mismatches in the Bayesian network
• Re-learn parts of the network

24
Life-long Adaptation

• Initial
• Error is detected and is localized
• Robot re-learns some local models

25
Life-long Adaptation
26
Evaluation

• Quick localization of error
• Robust recovery

27
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

28
Future Work

• Active self-exploration, optimal control, POMDPs
• Marker-less self-perception
• Moving robot
• Tool use