Robotic SelfPerception and Body Scheme Learning Jrgen Sturm
Robotic Self-Perception and Body Scheme Learning Jürgen Sturm Christian Plagemann Wolfram Burgard University of Freiburg Germany With annotated questions SA-1
Outline n Presentation of current research Clarification of Concepts, e. g. , - Mirror Neurons - Body Schema n Planned journal articles Possible support of our theory from a psychological point of view? n Future research / experiments Psychological evidence Co-experiments human/robot General Brainstorming
Motivation Existing robot models are typically n specified (geometrically) in advance n calibrated manually
Motivation Problems with fixed robot models: n Wear-and-tear wheel diameter, air pressure n Recovery from failure malfunctioning actuators n Tool use extending the model n Unknown model re-configurable robots
Motivation Problems with fixed robot models: n Wear-and-tear wheel diameter, air pressure n Recovery from failure malfunctioning actuators n Tool use extending the model n Unknown model re-configurable robots Similar problems in humans/animals?
Motivation Problems with fixed robot models: n Wear-and-tear wheel diameter, air pressure n Recovery from failure malfunctioning actuators n injured body parts Tool use extending the model n growth, aging writing Unknown model re-configurable robots riding a bike Similar problems in humans/animals?
Related Work n Neuro-physiology Mirror neurons n Body Schemes n n Clarification of concepts Better references? Good primer? [Rizzolatti et al. , 1996] [Maravita and Iriki, 2004] Robotics Self-calibration [Roy and Thrun, 1999] n Cross-modal maps [Yoshikawa et al. , 2004] n Structure learning [Dearden and Demiris, 2005] n Where else? E. g. , - Self-configuring software - Language acquisition
Problem motivation Fixed-model approaches fail when n parameters change over time n geometric model is not available Our Contribution n n 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 6 D Poses Motor babbling Act Joint angles
Problem Formulation n Visual self-perception of n body parts: X 1 ; : : : ; X n 2 R 4£ 4 n Actuators (m action signals): a 1 ; : : : ; am 2 R Learn the mapping Which brain area does this mapping? p (X 1 ; : : : ; X n ja 1 ; : : : ; am ) Body pose Configuration
Existing Methods n Analytic model + parameter estimation Requires prior knowledge n Function approximation n Nearest neighbor n Neural networks High-dimensional learning problem Requires large training sets
Body Scheme Factorization Local models similar to mirror neurons? We represent the kinematic chain as a Idea: Factorize the model 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 n n Using Gaussian process regression Learn 1 D 6 D transformation function p(¢ 12 j a 1 ) = p(X 1¡ 1 X 2 j a 1 ) for each (action, marker) triple
Finding the Network Structure n Select the most likely network topology n Corresponding to the minimum spanning tree Maximizing the data likelihood n p(M j. D)
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 n Purpose: n n prediction of end-effector pose in a given configuration Approach: n n n integrate over the kinematic chain in the Bayesian network by concatenating Gaussians approximate the result efficiently by one Gaussian p (X n j. X 1 ; a 1 ; : : : ; am ) = Z Z : : : p. M : : : d. X 2 ; : : : ; d. X n ¡ 1 2 1
Inverse Kinematics n n Purpose: Generate motor commands for reaching a given target pose Approach: Estimate Jacobian of endeffector using forward kinematics prediction · r X n (a) n = @X n (a) ; : : : ; @a 1 @am Use standard IK techniques n Jacobian pseudo-inverse ¸
Experiments
Evaluation: Forward Kinematics n n Fast convergence (approx. 10 -20 iterations) High accuracy (higher than direct perception)
Evaluation: Inverse Kinematics n Accurate control using bootstrapped body scheme
Life-long Adaptation Robot’s physical properties Physiology? Anatomy? will change over time n Predictive accuracy of body scheme needs to be monitored continuously n 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? n Initial n Error is detected and is localized n Robot re-learns some local models
Life-long Adaptation
Evaluation Recovery time plot for a human after body “deformation”? Quick localization of error n Robust recovery n
Summary n Novel approach learning body schemes from scratch using visual self-perception n n Model learning using Gaussian process regression Model selection using data likelihood as criterion n Efficient adaptation to changes in robot geometry n Accurate prediction and control
Future Work n Active self-exploration, optimal control, POMDPs n Marker-less self-perception n Moving robot n Tool use
Future work: Tool Use n Using tools requires dynamic extensions of the body scheme
Future research / experiments n Tool use n Writing with a pen n Approach: • • n 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) n Tutoring Evasion Maneuvers using Tactile Sensors (Frederic Dijoux)
Student projects (2) n Model-Free Control for Robotic Manipulators using Nearest-Neighbor methods (Hannes Schulz and Lionel Ott)
Student projects (3 a) n Dynamically adding repellant end-effectors (Clemens Eppner)
Student projects (3 b) n Programming by Demonstration (Clemens Eppner)
Student projects (3 c) n Programming by Demonstration (Clemens Eppner)
Student projects (3 d) n Programming by Demonstration (Clemens Eppner)
Student projects (4 a) n Object Recognition using Tactile Sensors (Alexander Schneider)
Student projects (4 b) n Object Recognition using Tactile Sensors (Alexander Schneider)
Student projects (5 a) n Grasping objects using Visual Servoing (Nikolas Engelhard) (Video courtesy of TU Dortmund)
Student projects (5 b) n Grasping objects using Visual Servoing (Nikolas Engelhard) (Video courtesy of TU Dortmund)
Planned journal article n Special Issue “Journal of Physiology” n n Neuro-Robotics Symposium – Sensorimotor Control, July 2008, Freiburg Two reviewers, one from neuro-biology, one from engineering, Deadline: 22. 10. 2008 Article Content: n Similar to this presentation n Stronger focus on mirror neurons and body schemas in humans/animals Support from psychological point of view?
Possible journal article n Special Issue “Autonomous Mobile Manipulation” n n Journal “Autonomous Robots” Deadline: 15. 12. 2008 Article Content (if at all): n Focus on Model selection? n. .
Brainstorming Psychological input n Co-experiments human/robot n Joint (student) project(s)? n. . n
- Slides: 44