1 / 43

Learning and Remapping in the Sensory Motor System

Learning and Remapping in the Sensory Motor System. F.A. Mussa-Ivaldi Northwestern University. In natural arm movements, complex motions of the joints lead to simple hand kinematics. From Morasso 1981. Is the shape of trajectories the byproduct of dynamic optimization?.

Télécharger la présentation

Learning and Remapping in the Sensory Motor System

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.


Presentation Transcript

  1. Learning and Remapping in the Sensory Motor System F.A. Mussa-Ivaldi Northwestern University

  2. In natural arm movements, complex motions of the joints lead to simple hand kinematics From Morasso 1981

  3. Is the shape of trajectories the byproduct of dynamic optimization? (Uno, Kawato and Suzuky, 1989)

  4. Flanagan and Rao (1995) Kinematic Adaptation

  5. Dynamic Adaptation Unperturbed movements Shadmehr & Mussa-Ivaldi, 1994

  6. Note: there is no force at rest Perturbing field

  7. After-effects Adaptation

  8. Dingwell, Mah and Mussa-Ivaldi (2004) Transport of a spring-mass object

  9. Baseline Initial response Post-adaptation kinematics

  10. The shape of trajectories is not a by-product of dynamic optimization. It reflects and is consistent with the Euclidean properties of ordinary space

  11. What is space ? • Intuitive Concept Geometry, from a physical standpoint is the totality of the laws according to which rigid bodies mutually at rest can be placed with respect to each other… “Space” in this interpretation is in principle an infinite rigid body – or skeleton - to which the position of all other bodies is related.” A. Einstein, 1949 • Less Intuitive Concepts Signals spaces, configuration spaces, state spaces, feature spaces, etc...

  12.  d y x Ordinary Space Two equivalent statements • Euclid • Pythagoras

  13. Euclidean (L2) Norm Orthogonal transformation Euclidean Symmetry Among all possible norm only the L2 (Euclidean) norm is invariant under rigid transformations (e.g. rotations, translations). Size does not depend upon position and orientation.

  14. Signal spaces • Neither the visual nor the motor signals spaces are Euclidean. • Yet, we perceive Euclidean symmetry and we coordinate motor commands to control independently movement amplitude and direction

  15. It is a common misconception that the visual system provides the primal information about this symmetry. • In fact it is likely for the opposite to be true: we learn the properties of space through active perception

  16. Is the bias toward Euclidean symmetry a guiding principle for the emergence of new motor programs?

  17. The motor system can reassign degrees of freedom to novel tasks Glove Talk. Sidney Fels and Geoffrey Hinton, 1990 Vocabulary Sam-I-am

  18. The properties of space are captured by the motor system when forming novel motor programs whose purpose is to control the motion of an endpoint

  19. Remapping Space 19 SIGNALS (Hand Configuration) 2 Screen Coordinates F.A. Mussa-Ivaldi, S. Acosta, R. Scheidt, K.M. Mosier

  20. The task • The subject must place the cursor corresponding to the hand inside a circular target • Experiment begins with the cursor inside a target. • New target is presented – Cursor is suppressed. • The subject must try to make a single rapid movement of the fingers so as to place the (invisible) cursor inside the new target and stop. • As the hand is at rest after this initial movement, the cursor reappears • The subject corrects by moving the cursor under visual guidance inside the target • Only the initial open loop movement is analyzed

  21. Some examples Part 2 Part 1 A B D C

  22. Some observations • Weird task. Cannot be represented before the experiment • Body and task endpoint are physically uncoupled • Only feedback pathway is vision • Dimensional imbalance • Metric imbalance

  23. Single Subject Ensemble Endpoint error

  24. Aspect-Ratio: AB/SE E Ensemble Single Subject B A S Linearity • Cursor movements DO NOT become straighter !

  25. Extended training • 4 consecutive days • 2 groups of subjects • No-Vision subjects (P2) just engage in the task for 4 days • Vision subjects (P3) train with continuous visual feedback trials. But they are tested in the same no-vision condition as the other group. • The two groups receive the same amount of training

  26. Extended Training Results • Both group have the same learning trend. • Continuous visual feedback of endpoint movements facilitates the formation of straighter open-loop trajectories

  27. Error reduction and straightening of motion are independent processes • Straightening of motion is facilitated when vision and movement execution are concurrent

  28. The subjects learn to solve an ill-posed inverse problem (from cursor location to hand gesture) and in doing so, they learn the geometrical structure of the map, i.e. a specific partition of degrees of freedom.

  29. Decomposition of Glove Signals

  30. Task-space and Null-space Motion ↓19% ↓24% ↓20% ↓23%

  31. Importing a metric structure

  32. Practical Applications:Body-Machine Interfaces (BMI) Interface • Identify Residual Control Signals (Movements, EEG, EMG, Multiunit…) • Form a first “vocabulary” of commands (faster, slower, right, left, MIDI, etc…) • Machine learns to respond appropriately • Patient practice and leans new commands DUAL LEARNING

  33. Sensing garments (Smartex -Pisa) • 52 smeared sensors on lycra • No metallic wires • External connection realized by metallic press stud

  34. Calibration: Find optimally controlled DOFs (synergies) by dimensionality reduction

  35. Natural Synergies -> Control Parameters A better map A bad map

  36. Tonal Spaces

  37. The interaction of human and machine learning: Co-Adaptation by Least Mean Squares

  38. The interactive LMS procedure - 1 1)Set an arbitrary starting map: Endpoint vector dim(x) =2 “Body” vector dim(h)=20

  39. The interactive LMS procedure - 2 2) Collect movements toward a set of K targets: Targets Movements (endpoint coordinates) Movements (body coordinates)

  40. The interactive LMS procedure - 3 3) Calculate the error (in endpoint coordinates) For each movement For a whole test epoch

  41. The interactive LMS procedure - 4 4) Calculate the gradient of the error (w.r.t. A) 5) Update the A-matrix in the direction of the gradient 4) Repeat from step 2 until convergence.

  42. An Example

  43. Conclusion Sensory-Motor Cognition IS NOT an oxymoron !

More Related