Anshul Joshi

Robot Cognition using Bayesian Symmetry Networks

(Leyton, 2001) proposed a generative theory of shape, and general cognition, based on group actions on sets as defined by the wreath product. Our position expressed here is that this approach can provide a strong basis for robot cognition when: 1. tightly coupled to sensorimotor data and analysis, 2. used to structure both general concepts and specific instances, and 3. combined with a probabilistic framework (Bayesian networks) to characterize uncertainty. We describe a roadmap to achieve these and provide some evidence of feasibility.

Bayesian Computational Sensor Networks: Small-scale Structural Health Monitoring☆

Abstract The Bayesian Computational Sensor Network methodology is applied to small-scale structural health monitoring. A mobile robot, equipped with vision and ultrasound sensors, maps small-scale structures for damage (e.g., holes, cracks) by localizing itself and the damage in the map. The combination of vision and ultrasound reduces the uncertainty in damage localization. The data storage and analysis takes place exploiting cloud computing mechanisms, and there is also an off-line computational model calibration component which returns information to the robot concerning updated on-board models as well as proposed sampling points. The approach is validated in a set of physical experiments.

Symmetry based semantic analysis of engineering drawings

Engineering drawings have posed significant challenges to image analysis for many decades. The goal is to take images of scanned engineering drawings and interpret them so as to understand their contents (e.g., characters, digits, line segments, box segments etc.). This is known as semantic analysis. We propose a new approach here which takes advantage of the man-made nature of drawings: there is a tremendous amount of symmetry. We exploit this insight to enhance our previously reported system, the Non-Deterministic Agent System (NDAS), with symmetry-based analysis tools. Agents work independently but use each others results to produce the final result (e.g., form segmentation, character analysis, structural analysis, boundary segmentation, etc.). We use the wreath product representation both to characterize symmetry as well as to structure a Bayesian network model of the uncertainty. This approach permits wide application to perform semantic analysis of engineering drawings.

Wreath product cognitive architecture (WPCA)

A Belief-Desire-Intention (BDI) framework closely resembles human practical reasoning approach in day-to-day life, and is a well-studied architecture. The wreath product cognitive model, first described by Leyton is an abstract, although powerful, model which closely couples perception and actuation for representing shape. However, no implementation of the wreath product model exists. Our work is an attempt to combine the wreath product knowledge representation mechanism with a BDI architecture that works in a real-world setting. A prototype implementation of this combination is demonstrated on an iRobot Create differential-drive robot, with a Kinect One structural sensor, in an indoor environment. The effectiveness of our framework is demonstrated by its accuracy for mapping the environment and localization of the robot for navigation purposes.

Actuation-based Shape Representation Applied to Engineering Document Analysis

We propose that human generated drawings (including text and graphics) can be represented in terms of actuation processes required to produce them in addition to the visual or geometric properties. The basic theoretical tool is the wreath product introduced by Leyton (Leyton, 2001) (a special form of the semi-direct product from group theory which expresses the action of a control group on a fiber group) which can be used to describe the basic strokes used to form characters and other elements of the drawing. This captures both the geometry (points in the plane) of a shape as well as a generative model (actuation sequences on a kinematic structure). We show that this representation offers several advantages with respect to robust and effective semantic analysis of CAD drawings in terms of classification rates. Document analysis methods have been studied for several decades and much progress has been made; see (Henderson, 2014) for an overview. However, there are many classes of document images which still pose serious problems for effective semantic analysis. Of particular interest here are CAD drawings, and more specifically sets of scanned drawings for which either the electronic CAD no longer exists, or which were produced by hand. We demonstrate results on a set of CAD-generated drawings for automotive parts.

Actuation in perception: Character classification in engineering drawings

We propose a novel approach to 2D character recognition by incorporating actuation data into the shape representation. Sensorimotor data is analyzed in terms of actuation sequences which generate the data. We illustrate the use of Wreath Products (WPs) to represent robot sensorimotor experience in a way that ties together perception and actuation. WPs naturally represent not only the Euclidean symmetries possessed by an object, but also the sequence of actuations used to generate those. Two distinct approaches using actuation signals to represent shape are compared: (1) the Kullback-Leibler measure is applied to histograms of translation symmetries in the shape, and (2) a distance metric is defined on pure actuation signals. Experimental results show that these methods achieve excellent classification rates (99 %) on text extracted from scanned images of engineering drawings for the top five hypotheses.