1 / 14

From motor babbling to hierarchical learning by imitation: a robot development pathway

From motor babbling to hierarchical learning by imitation: a robot development pathway. Yiannis Demiris and Anthony Dearden. By James Gilbert. Overview. Social vs. A-Social Learning Forward & Inverse Models HAMMER Architecture. Fundamental Questions.

ban
Download Presentation

From motor babbling to hierarchical learning by imitation: a robot development pathway

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. From motor babbling to hierarchical learning by imitation: a robot development pathway YiannisDemiris and Anthony Dearden By James Gilbert

  2. Overview • Social vs. A-Social Learning • Forward & Inverse Models • HAMMER Architecture

  3. Fundamental Questions • ‘How does an individual use the knowledge acquired through self exploration as a manipulable model through which to understand others and benefit from their knowledge?’ • ‘How can developmental and social learning be combined for their mutual benefit? ‘

  4. A-Social Learning • The robot attempts to learn to perform a task by interacting with its environment • Doesn’t consider the solutions that other agents in its environment possess with respect to the task.

  5. Social Learning • Learning by: • Observation, • Demonstration and, • Imitation. • Equipping robots with the ability to imitate enables them to learn to perform tasks by observing a human demonstrator (Schaal, 1999).

  6. Forward Model • ‘We aim at building a system that enables a robot to autonomously learn a forward model with no a priori knowledge of its motor system or the external environment.’ • The robot sends out random commands to its motor system (babbles), • Then together with the information from the vision system learns the structure and parameters of a Bayesian belief network.

  7. Inverse Model • Inversely mapping the learnt forward models gives a corresponding inverse model. • This allows the model to seek meaning from a subject being imitated by being able to ask itself what it would do in that situation.

  8. Coupled Model • By achieving a coupled inverse and forward model, a robot is capable of • Executing an action, • Rehearsing an action, or • Perceiving an action

  9. HAMMER Architecture • (HAMMER – Hierarchical Attentive Multiple Models for Execution and Recognition). • Attempts to combine social and a-social models. • ‘Adopts the simulation theory of mind approach to understand the actions of others, using a distributed network of coupled inverse and forward models.’ • New models can be added either through self-exploration, or by observing others, through imitation.

  10. HAMMER Architecture ctd. • The fundamental structure of HAMMER is an inverse model paired with a forward model. • The inverse model receives information about the current state, • Outputs the motor commands it believes are necessary to achieve target goals. • The forward model provides an estimate of the next state. • This is fed back to the inverse model, allowing it to adjust any parameter of the behaviour.

  11. HAMMER Architecture ctd. • Matching a visually perceived demonstrated action with the imitator’s equivalent one. • Feeding the demonstrator's current state as perceived by the imitator to the inverse model. • Generate the motor commands that it would output if it was in that state and wanted to execute an action. • The forward model outputs the estimated next state, which is a prediction of what the demonstrator’s next state will be. • The prediction is compared with the demonstrator’s actual state at the next step. • The comparison error can then be used for learning.

  12. Open Challenge • Forward models are learnt between the motor commands and the image plane based visual perception of the effects of these actions. A generalisation of these to more abstract representations (for example a 3-D body schema ) will allow more complex relationships to be addressed.

  13. Conclusion & Personal Thoughts • By using Bayesian Belief Networks to create both forward and inverse models, it was shown through a simple practical experiment that a robot can mimic the actions of a human. • However the robot doesn’t necessarily understand the meaning of the action. • The experimental robot was very simple and shares very little with the human form it was to mimic.

  14. Questions…

More Related