Wataru Takano

Incremental Learning, Clustering and Hierarchy Formation of Whole Body Motion Patterns using Adaptive Hidden Markov Chains

This paper describes a novel approach for autonomous and incremental learning of motion pattern primitives by observation of human motion. Human motion patterns are abstracted into a dynamic stochastic model, which can be used for both subsequent motion recognition and generation, analogous to the mirror neuron hypothesis in primates. The model size is adaptable based on the discrimination requirements in the associated region of the current knowledge base. A new algorithm for sequentially training the Markov chains is developed, to reduce the computation cost during model adaptation. As new motion patterns are observed, they are incrementally grouped together using hierarchical agglomerative clustering based on their relative distance in the model space. The clustering algorithm forms a tree structure, with specialized motions at the tree leaves, and generalized motions closer to the root. The generated tree structure will depend on the type of training data provided, so that the most specialized motions will be those for which the most training has been received. Tests with motion capture data for a variety of motion primitives demonstrate the efficacy of the algorithm.

Online Segmentation and Clustering From Continuous Observation of Whole Body Motions

This paper describes a novel approach for incremental learning of human motion pattern primitives through online observation of human motion. The observed time series data stream is first stochastically segmented into potential motion primitive segments, based on the assumption that data belonging to the same motion primitive will have the same underlying distribution. The motion segments are then abstracted into a stochastic model representation and automatically clustered and organized. As new motion patterns are observed, they are incrementally grouped together into a tree structure, based on their relative distance in the model space. The tree leaves, which represent the most specialized learned motion primitives, are then passed back to the segmentation algorithm so that as the number of known motion primitives increases, the accuracy of the segmentation can also be improved. The combined algorithm is tested on a sequence of continuous human motion data that are obtained through motion capture, and demonstrates the performance of the proposed approach.

Humanoid Robot's Autonomous Acquisition of Proto-Symbols through Motion Segmentation

Mimesis is the theory that human intelligence originated in the interactive communication of motion recognition and generation through imitation. A mimesis model has been proposed using hidden Markov models (HMMs), which represent proto symbols. In our previous system, the user had to manually divided a sequence of motion into segments in order to embed each segment as an HMM. Automatic segmentation is essential for a system to autonomously learn and develop through imitation. In this paper, we propose an automatic motion segmentation method utilizing correlation among movements for a short time period. In addition, we show that it is possible to acquire proto symbols by providing the automatically segmented motion patterns with the mimesis system

Primitive communication based on motion recognition and generation with hierarchical mimesis model

Communication skill is essential for social robots in various environments such as homes, offices, and hospitals, where the robots are expected to interact with humans. In this paper, we model the primitive nonverbal communication between two persons by mimetic communication model. The model consists of three groups of hidden Markov models (HMMs) hierarchically combined to recognize motions of the human and to generate the interactive motions of the robot. HMMs in the lower layer abstract the motion patterns and HMMs in the upper layer represent the interaction patterns. We demonstrate the validity of this model through kick boxing match between a motion-captured human and humanoid robot, where the robot can autonomously generate its motion in response to attacks by the human

Recognition of human driving behaviors based on stochastic symbolization of time series signal

This paper describes an imitative learning of driving time series data for intellectual cognition toward future automobiles. The driving pattern primitives consisting of states of the environment, vehicle and driver are symbolized by hidden Markov models (HMMs), which can be used for both recognition and generation of the driving patterns. The relationship among the HMMs can be represented by locating the HMMs in a multidimensional space. The contribution of each variable to the HMM space can be analyzed such that important variables can be selected out of the driving data in order to reduce the size of the HMMs. Moreover, this paper presents a hierarchical model with the HMMs abstracting the primitive driving patterns in the lower layer, and another HMMs abstracting the longterm contextual driving patterns which are representation in the HMM space. Tests with a driving simulator and a actual vehicle demonstrate not only the validity of symbolization of driving pattern primitives, recognition and generation, but also availability of key feature selection. The extended hierarchical model is also proved to have a potential to predict the driving data appropriately.

Representability of human motions by factorial hidden Markov models

This paper describes an improved methodology for human motion recognition and imitation based on factorial hidden Markov models (FHMM). Unlike conventional hidden Markov models (HMMs), FHMMs use a distributed state representation, which allows for more efficient representation of each time sequence. Once the FHMMs are trained with exemplar motion data, they can be used to generate sample trajectories for motion production, and produce significantly more accurate trajectories compared to single Hidden Markov chain models. Due to the additional information encoded in FHMMs models, FHMM models have a higher Kullback-Leibler distance compared to single Markov chain models, making it easier to distinguish between similar models. The efficacy of using FHMMs is tested on a database of human motions obtained through motion capture. The results show that FHMMs provide better generalization to new data when compared to conventional HMMs during motion recognition, as well as providing a better fit for generated data.

Incremental on-line hierarchical clustering of whole body motion patterns

This paper describes a novel algorithm for autonomous and incremental learning of motion pattern primitives by observation of human motion. Human motion patterns are abstracted into a hidden Markov model representation, which can be used for both subsequent motion recognition and generation, analogous to the mirror neuron hypothesis in primates. As new motion patterns are observed, they are incrementally grouped together using hierarchical agglomerative clustering based on their relative distance in the HMM space. The clustering algorithm forms a tree structure, with specialized motions at the tree leaves, and generalized motions closer to the root. The generated tree structure will depend on the type of training data provided, so that the most specialized motions will be those for which the most training has been received. Tests with motion capture data for a variety of motion primitives demonstrate the efficacy of the algorithm.

Capture Database through Symbolization, Recognition and Generation of Motion Patterns

Motion capture systems are used to obtain motion data such that humanoid robots or computer graphics (CG) characters can behave naturally. However, it has proven to be hard not only to modify the capture data without losing its reality but also to search for the required capture data in a lot of capture data. In this paper, we provide a solution to these problems based on our previous work on symbolization of motion patterns for developing humanoid intelligence. Similar motion sequences in the database are abstracted as a symbol, which will be applied to searching motion patterns in the database similar to a given motion. This paper also introduces a method for building a stochastic symbol-word mapping model utilizing the word labels provided by the operator during motion capture sessions. This model converts a input sequence of words into a sequence of symbols, and then allows the capture database both to be searched for capture data corresponding to the input (a sequence of words) and to provide the users with new motion data generated by the symbols. Finally, we apply analogy of symbols to establishing the database in order to provide an appropriate motion data in response to an unsupervised sequence of words and then demonstrate the validity of analogy theory.

Statistically integrated semiotics that enables mutual inference between linguistic and behavioral symbols for humanoid robots

This paper describes the linguistic model based on symbolization of motion patterns for humanoid robots. The model consists of two kinds of stochastic models : the motion language model and the natural language model. The motion language model stochastically connects the symbols of motion patterns to the morpheme words through the latent states which represent the underlying linguistic structure such as semantic contents. The natural language model represents the dynamics of the word classes. The motion language model and the natural language model correspond to semantics and syntax respectively. The integration of the motion language model and the natural language model allows robots not only to linguistically interpret the motion patterns as sentences but also to generate the motions from the sentences. The two kinds of linguistic processes of the interpretation and the generation can be obtained by solving search problems: search for a sequence of morpheme words and a symbol of motion pattern. The proposed approach to interpretation of motion patterns as sentences and generation of motion patterns from the sentences through integration of the motion language model and the natural language model is validated by the experiment on the human behavioral data.

Statistical mutual conversion between whole body motion primitives and linguistic sentences for human motions

This paper describes a novel approach to linguistic mutual inference, which enables robots not only to linguistically interpret the motion patterns in the form of sentences but also to generate the motions from the sentences. The inference can be established based on two modules, the motion language model and the natural language model. The motion language model stochastically represents an association structure between symbols of motion patterns and the words in sentences assigned to the motion. This is a statistical model with a three layered structure of motion symbols, latent states and words. The natural language model statistically represents a structure of sentences based on word bigrams. The motion language model and the natural language model correspond to semantics and syntax respectively. An approach to the integration of motion language model with the natural language model allows the linguistic mutual inference for the robots. The two kinds of inference can be made by solving search problems, search for a sequence of words corresponding to a motion and search for a symbol of motion pattern corresponding to a sentence. The proposed approach to interpretation of motion patterns as sentences and generation of motion patterns from the sentences through the integration of motion language model with the natural language model is validated by an experiment on the human behavioral data.