Object manifold learning with action features for active tactile object recognition

Abstract

Abstract-In this paper, we consider an object recognition problem based on tactile information using a robot hand. The robot performs an exploratory action to the object to obtain the tactile information, however, poorly designed actions may not be sufficiently informative. In contrast, if we could collect sample data by sequentially performing informative actions, i.e., active learning, the required time would be drastically reduced. To this end, we propose a novel approach for active tactile object recognition. Our approach combines both an active learning scheme and a nonlinear dimensionality reduction method. We first extracts the object manifold, each coordinate of which represents an object, from tactile sensor data and action features using Gaussian Process Latent Variable Models. At the same time, a probabilistic model of the observed data related to the action and the object are learned. Then, with the learned model, optimally-informative exploratory actions can be computed sequentially, and performed to efficiently collect the data for recognition. We show experimental results that verify the effectiveness of our proposed method with synthetic data and a real robot.

Active Tactile Transfer Learning for Object Discrimination in an Unstructured Environment Using Multimodal Robotic Skin

“…However, the experiments were only carried out in the simulation using uniformly collected data offline. Tanaka et al 36 combined Gaussian process latent variable and nonlinear dimensionality reduction method to actively discriminate among four cups in the real experiments. The authors collected 400 training samples uniformly using three fingers of the Shadow hand, which was fixed and the objects were placed on a turntable.…”

Contact localization through spatially overlapping piezoresistive signals

“…Bekiroglu et al [31] also studied how grasp stability can be assessed based on tactile sensory data using machine-learning techniques like AdaBoost, SVM and HMM with similar success. In similar fashion, both Saal et al [32] and Tanaka et al [33] used probabilistic models on tactile data to estimate object dynamics and perform object recognition respectively. However, most of this work is based on tactile arrays built on rigid substrates and thus unable to provide full coverage of complex geometry.…”

Bayesian Action&Perception: Representing the World in the Brain

“…Tanaka et al recently took advantage of its ability to learn distinguishing features empirically (Tanaka et al, 2014) rather than requiring a priori parameterization (Saal et al, 2010). It can also be applied to autonomous, open-ended classification without requiring supervised learning of a specific database (Dallaire et al, 2014).…”