1 / 17

Resilient Machines Through Continuous Self-Modeling

Resilient Machines Through Continuous Self-Modeling. Josh Bongard,Victor Zykov , and Hod Lipson, Science, Vol.314, pp. 1118-1121, 2006. Pattern Recognition 2010.04.06 Seung -Hyun Lee Soft Computing Lab. Contents. Introduction Motivation Self Modeling Experiments Conclusion.

Download Presentation

Resilient Machines Through Continuous Self-Modeling

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.

E N D

Presentation Transcript


  1. Resilient Machines ThroughContinuous Self-Modeling Josh Bongard,VictorZykov, and Hod Lipson, Science, Vol.314, pp. 1118-1121, 2006. Pattern Recognition 2010.04.06 Seung-Hyun Lee Soft Computing Lab.

  2. Contents • Introduction • Motivation • Self Modeling • Experiments • Conclusion

  3. Introduction • Animals • After injured, create qualitatively different compensatory behaviors • Robots • How robots can deal with this sort of unexpected damage?  self modeling

  4. Motivation • How can robot learn its own morphology? • Direct observation? • Database of past experience? • How can robot synthesize complex behaviors or recover from damage? • Trial and error?  slow, costly, risky! • In this paper, • Inferring morphology: self-directed exploration • Complex behavior or recovering from damage: synthesize new behaviors using the resulting self models

  5. Overall Process Self Modeling Prediction Modeling Testing

  6. Testing Self Modeling • In this process • Performs an arbitrary motor action • Records the resulting sensory data

  7. Modeiling Self Modeling • Model synthesize component • Synthesizes a set of candidate self-models • Method • Before damage(topological modeling) • Greedy random-mutation hill climber algorithm • 16 parameters Robot initially knows how many body pars it is composed of, the size, weight and mass of each part, and angle-movement relations • 15 random models • 200 iterations • Evaluation: Euclidean distance between the centroid and where the centroid should be • After damage(parametric modeling) • Self-model is frozen • 8 parameters (volumes and masses are scaled by 10%~200%)

  8. Prediction Self Modeling • Action synthesize component • Find a new action most likely to elicit the most information from the robot based on the current self model inferred

  9. Self Modeling • After self modeling procedures(16 times repetition) • Create desired behaviors (D) • Execute by the physical robot

  10. Self Modeling

  11. Robot Experiments • Speculation • 4 upper and lower leg parts and a main body • 8 motorized joints(-90 ~ 90 degree range) • 0 degree: flat • Positive degree: upwards • Negative degree: downwards • 2 tilt sensors • Self model representation • Planar topological arrangement • Damage • Disabled one leg

  12. Design Experiments • Control variables • Computational efforts(250,000 internal model simulations) • Physical actions(16) • Three algorithms • Algorithm 1: 16 random physical actions  batch training(modeling) • Algorithm 2: Physical actions  self modeling  random action selection • Algorithm 3(proposed): Physical actions self modeling  actions selection

  13. Result Experiments

  14. Result Experiments • Model-driven algorithm is more accurate than random baseline algorithms • A robot that actively chooses action on the basis of its current set of hypothesized self-models has a better chance of successfully inferring its own morphology

  15. Result Experiments • Automatically generated self-model was sufficiently predictive to allow the robot to consistently develop forward motion patterns without further physical trials

  16. Result Conclusion • Contribution • First physical system • Autonomously recover its own morphology with little prior knowledge • Optimize the parameters of its morphology after unexpected change • Show the possibility of unknown cognitive process • Which organisms actively create and update self models in the brain? • How and which sensor-motor signals are used to do this? • What form these model take? • Does human utilize multiple competing models?

  17. Thank you

More Related