Chapter 11. A Bayesian model of imitation in infants and ro - PowerPoint Presentation

Slide1

Chapter 11. A Bayesian model of imitation in infants and robots (2/2)in Imitation and Social Learning in Robots, Humans and Animals, Nehaniv & DautenhahnCourse: Robots Learning from Humans

Dong-Kyoung Kye2015. 11. 13Vehicle Intelligence LaboratorySchool of Electrical and Computer EngineeringSeoul National Universityhttp://vi.snu.ac.kr

VEHICLE INTELLIGENCE LABSlide2

ContentsBayesian imitative learningExample : learning to solve a maze task through imitationLearning a forward model for the maze taskImitation using the learned forward model and learned priorsInferring the intent of the teacherFurther applications in robotic learningTowards a probabilistic model for imitation in infantsConclusion2Slide3

Bayesian imitative learning - Forward model3It maps state, action to next state :

Learned from exploring the state-space at random

Body babbling

Supervised process (Assuming proprioception)

 Slide4

Bayesian imitative learning - Inverse model4Probability that an action is chosen given the desired next state, and the goal

Forward model

PriorSlide5

Learning to solve a maze task through imitation5Learning a forward model for the maze task20 x 20 grid of squaresStates

: Grid locations in the mazeFive actions available - North(N), East(E), South(S), West(W) or remain in place(X)The noisy ‘forward dynamics’ of the environment 

- Simulated maze environment

- Actual and learned probabilistic forward modelsSlide6

Learning to solve a maze task through imitation6Imitation using the learned forward model and learned priorsThe imitator can use ‘inverse model’ to select appropriate actions to imitate the teacher and reach the goal state.

Forward model

Inverse model

PriorSlide7

Learning to solve a maze task through imitation7Imitation using the learned forward model and learned priorsThe imitator can use ‘inverse model’ to select appropriate actions to imitate the teacher and reach the goal state.

- Simulated maze environment

PriorSlide8

Learning to solve a maze task through imitation8Imitation using the learned forward model and learned priorsSlide9

Learning to solve a maze task through imitation9Inferring the intent of the teacherThe intent inference algorithm provides an estimate of the distribution over the instructor’s possible goals for each time step.

Learned

forward model

Learned prior over actionsSlide10

Learning to solve a maze task through imitation10With a maze task example.. It shows that how the abstract probabilistic framework proposed in this chapter can be used to solve a concrete sensorimotor problem.Slide11

Further applications in robotic learning11E.g.) Box lifting (HOAP-2)Slide12

Further applications in robotic learning12E.g.) Box lifting (HOAP-2)Learning forward models from motion capture.(a) Forward models learned by the system after observing 3 different actions performed by the human(b) Forward model learned by the system for the box lift experimentSlide13

Further applications in robotic learning13E.g.) Box lifting (HOAP-2)Slide14

Towards a probabilistic model for imitation in infants14- Active Intermodal Mapping (AIM)Slide15

Towards a probabilistic model for imitation in infants15- Match and correction process is the Bayesian action selection method.Slide16

ConclusionBayesian approach is well-suited to imitation learning in real-world robotic environments which are noisy and uncertainBayesian probabilistic framework can also be applied to better understand the stages of infant imitation learning.16