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

Abstract

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.

Estimating the non-linear dynamics of free-flying objects

“…[38]. The Hidden Markov model (HMM) is used for recognition and regeneration of human motion across various demonstrations [39,40], and to teach a robot to perform assembly tasks [41]. Locally Weighted Projection Regression (LWPR) is used to approximate the dynamic model of a robot arm for computed torque control [42,32], for teaching a robot to perform basic soccer skills [43], and is also applied for real-time motion learning for a humanoid robot [32].…”

Incremental learning of human behaviors using hierarchical hidden Markov models

“…In previous work, we have been developing algorithms for autonomous segmentation and extraction of movement primitives from continuous observation of human motion [2], [3]. Using continuous time-series data as the input, we first segment the data into potential motion primitives, using a modified version of the Kohlmorgen and Lemm [24] algorithm for unsupervised segmentation.…”

“…In addition, in order to be applicable in arbitrary human domains, where the task and the demonstrator may be changing, the learning system should be capable of continuous learning of both primitives and their sequencing, during online observation of human motion. In previous work [2], [3], we have been developing algorithms for incremental learning of movement primitives from continuous observation, based on stochastic modeling. In this paper, we propose an approach for simultaneously also learning the higher level order in the sequencing, by extending the stochastic model to higher levels of abstraction.…”

Comparative study of representations for segmentation of whole body human motion data

“…Note that this algorithm models the entire segment as a single state (single probability distribution), and is therefore not a generative model that can be used to generate simulations of the derived segments. Once motion segments are extracted, they can be incrementally clustered and modeled with more detailed stochastic models to produce generative models for use during motion synthesis [23].…”