Mutual Development of Behavior Acquisition and Recognition based on Value System

Takahashi, Yasutake; Tamura, Yoshihiro; Asada, Minoru

doi:10.3156/jsoft.21.640

Cited by 2 publications

(3 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…From the view point of reinforcement learning framework, Meltzoff's "Like me" hypothesis indicates that a child estimates rewards received by his/her caregiver based on experiences of self. Estimation of rewards received during the observed behavior based on reward models of own behavior enables an agent to understand/recognize/learn the observed behavior deeply [6], [7]. View estimation is important to understand/imitate the observed behavior because the mapping from sensory information to motor skill would be based on an egocentric coordinate, not on an allocentric one.…”

Section: Discussionmentioning

confidence: 99%

“…From a viewpoint of reinforcement learning framework, this hypothesis indicates that the reward and state value of the demonstrator might be estimated through observing the behavior. Takahashi et al proposed a method to understand observed behavior based on the state value estimation [6] and a method for mutual development of acquisition and recognition of observed behavior [7]. From the above viewpoints, the TD error might be utilized not only for behavior learning but also for the view estimation.…”

Section: Introductionmentioning

confidence: 98%

See 1 more Smart Citation

View estimation learning based on value system

Takahashi

Shimada

Asada

2009

2009 IEEE International Conference on Fuzzy Systems

Self Cite

View full text Add to dashboard Cite

Estimation of a caregiver's view is one of the most important capabilities for a child to understand the behavior demonstrated by the caregiver, that is, to infer the intention of behavior and/or to learn the observed behavior efficiently. We hypothesize that the child develops this ability in the same way as behavior learning motivated by an intrinsic reward, that is, he/she updates the model of the estimated view of his/her own during the behavior imitated from the observation of the behavior demonstrated by the caregiver based on minimizing the estimation error of the reward during the behavior. From this view, this paper shows a method for acquiring such a capability based on a value system from which values can be obtained by reinforcement learning. The parameters of the view estimation are updated based on the temporal difference error (hereafter TD error: estimation error of the state value), analogous to the way such that the parameters of the state value of the behavior are updated based on the TD error. Experiments with simple humanoid robots show the validity of the method, and the developmental process parallel to young children's estimation of its own view during the imitation of the observed behavior demonstrated by the caregiver is discussed.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 98%

View estimation learning based on value system

Takahashi

Shimada

Asada

2009

2009 IEEE International Conference on Fuzzy Systems

Self Cite

View full text Add to dashboard Cite

show abstract

“…The observed behaviours are chosen with respect to the similarity to the current situation. In [19] it is shown that not only the recognition of the observed action is important, but also to know the value of it. To acquire information from a demonstrator does not mean to replicate his movements, but to understand his intention and find a way to reach the same goal.…”

Section: Related Workmentioning

confidence: 99%

Increasing the Autonomy of Mobile Robots by Imitation in Multi-robot Scenarios

Richert

Scheller

Koch

et al. 2009

2009 Fifth International Conference on Autonomic and Autonomous Systems

View full text Add to dashboard Cite

Because imitation has shown to be able to drastically cut down the exploration space it has been embraced largely by the robotics community to speed up autonomous learning. This has been done mostly in fixed demonstratorimitator relationships, where a predefined demonstrator is performing the same action over and over again until the robot has learned it sufficiently. In practical multi-robot scenarios, however, the imitation process should not interrupt the observed robot as it will affect its autonomy. Therefore the imitating robot often has only a limited amount of behaviors of the same type to learn from. As this usually does not provide enough information for learning a generalized version of observed lowlevel actions, it can help the observer to optimize its strategy based on the recognized actions. In this case, first of all the imitating robot has to interpret the observed behavior in terms of its own behavior knowledge. This means that low-level observations have to be segmented into episodes of apparently similar behavior. For these episodes corresponding skills in the observing robot's own behavior repertoire have to be found. And for the episodes sequence corresponding state changes in the observing robot's own strategy have to be determined. How the former can be done has been shown by the authors in a previous paper. In this paper we will describe how our approach for Evolving Societies of Learning Autonomous Systems (ESLAS) is extended to be capable of multi-robot imitation. We demonstrate how the recognition results can be used in a multi-robot scenario to decentrally align the robots' behaviors, to speed up the overall learning process, and thus increase the overall autonomy.

show abstract

Mutual Development of Behavior Acquisition and Recognition based on Value System

Cited by 2 publications

References 25 publications

View estimation learning based on value system

View estimation learning based on value system

Increasing the Autonomy of Mobile Robots by Imitation in Multi-robot Scenarios

Contact Info

Product

Resources

About