Simple Kinesthetic Haptics for Object Recognition

TL;DR

This work tackles object recognition without tactile sensors by leveraging kinesthetic glances and a frame-invariant grasp representation learned from CAD models. It introduces the injective map $\Phi_n$ that projects a grasp to a $w$-dimensional vector $\mathbf{q}$ and two inference schemes, Iterative Classification (IC) and Sequential Bayesian Update (BC), to accumulate evidence across multiple grasps. A scale-invariant grasp representation is developed via normalization by $A_v$ and extended to $z$-finger grasps to increase information content. Experiments on KIT and GraspNet datasets, plus real and simulated grippers on YCB objects, show high recognition accuracy, scalability across sizes, and practical data-efficiency for hand-agnostic robotic systems.

Abstract

Object recognition is an essential capability when performing various tasks. Humans naturally use either or both visual and tactile perception to extract object class and properties. Typical approaches for robots, however, require complex visual systems or multiple high-density tactile sensors which can be highly expensive. In addition, they usually require actual collection of a large dataset from real objects through direct interaction. In this paper, we propose a kinesthetic-based object recognition method that can be performed with any multi-fingered robotic hand in which the kinematics is known. The method does not require tactile sensors and is based on observing grasps of the objects. We utilize a unique and frame invariant parameterization of grasps to learn instances of object shapes. To train a classifier, training data is generated rapidly and solely in a computational process without interaction with real objects. We then propose and compare between two iterative algorithms that can integrate any trained classifier. The classifiers and algorithms are independent of any particular robot hand and, therefore, can be exerted on various ones. We show in experiments, that with few grasps, the algorithms acquire accurate classification. Furthermore, we show that the object recognition approach is scalable to objects of various sizes. Similarly, a global classifier is trained to identify general geometries (e.g., an ellipsoid or a box) rather than particular ones and demonstrated on a large set of objects. Full scale experiments and analysis are provided to show the performance of the method.

Figures (21)

• Figure 1: Robotic haptic glances to identify (left) an occluded object in a cabinet and (right) on a table.
• Figure 2: Illustration of the (top) training and (bottom) evaluation processes. CAD models of the objects are used to generate labeled data for training a classifier. The classifier, trained solely over CAD models and independent of any particular robot hand, can now be used to classify real object. Recognition of real objects is done by random sampling grasps using a robotic hand in an iterative classification process - Iterative Classification (IC) or Bayesian Classification (BC).
• Figure 3: Example of a 4-vertex polygon describing a 4-finger ($n=4$) grasp of object $\mathcal{O}$ with a vector of nine parameters.
• Figure 4: Example of a 4-finger grasp of a box. A 4-vertex polyhedron formed by four contact points is defined by six parameters. The contact normals are marked in red while their eight parameterization angles are omitted for simplicity.
• Figure 5: Eight objects from the KIT object models database used in the experiments.

Theorems & Definitions (11)

• Theorem 1
• proof
• Definition 1
• Theorem 2
• proof
• Definition 2
• Definition 3
• Theorem 3
• proof
• proof