Mapping with Sparse Local Sensors and Strong Hierarchical Priors

Abstract

Abstract-The paradigm case for robotic mapping, as in Simultaneous Localisation and Mapping problems, considers a mobile robot with noisy odometry and laser scanners. Laser scanners provide large amounts of sensory information, and have effectively unlimited range in indoor environments. Such large quantities of input information allow the use of relatively weak priors. In contrast, the present study considers the mapping problem in environments where only sparse, local sensory information is available. To compensate for the lack of likelihood evidence, we make use of strong hierarchical object priors. Hierarchical models were popular in classical blackboard systems but hare here applied in a Bayesian setting and novelly deployed as a mapping algorithm. We give proof of concept results, intended to demonstrate the algorithm's applicability as a part of a tactile SLAM module for the whiskered ScratchBot mobile robot platform. I. INTRODUCTIONThe paradigm case for robotic mapping, as in Simultaneous Localisation and Mapping (SLAM) problems [1], considers a mobile robot with noisy odometry and SICK laser scanners. Laser scanners provide large amounts of sensory information, and have effectively unlimited range in indoor environments. Such large quantities of input information allow the use of relatively weak priors, such as independent grid cell occupancy or flat priors over the belief of small feature sets [1].In contrast, the present study considers the mapping problem in environments where only sparse, local sensory information is available. For example, a fire-fighting robot building up a map in a smoke-filled house cannot rely on laser scanners functioning at all times, and could instead operate by feeling its way around with touch sensors. Proof that this type of navigation is possible is found in biology: electric fish make use of highly localised electric field sensors [2] and rats navigate through dark underground tunnels using their whiskers [3], [4], both having ranges of a few centimetres. In robotics, touch sensors are relatively cheap in both material and computational processing terms, and their use has previously been considered to enhance navigation in cheap household robots [5]

CrunchBot: A Mobile Whiskered Robot Platform

“…1(a)) which draws together several strands of research towards systems for autonomous whiskerbased tactile navigation. We and other authors have previously investigated individual components of such a system in isolation, including whiskered texture recognition [4], [11], [2], [5], [13], surface shape recognition [12], [9], [8], [3], and object recognition [7]. These components are often tested under ideal laboratory conditions or in individual mobile settings [16].…”

“…The present navigation method is driven by texture and distance observations only. An intermediate object recognition step could also be performed, as described in a companion paper [7].…”

Tactile SLAM with a biomimetic whiskered robot

“…However we have presented the first results of whiskered grid mapping, which may serve as a baseline for future, improved whiskered systems. For example, [11] presents an alternative, more complex whiskered object recognition system, which could in future form a navigation component, and it would be useful to compare its results with the baseline presented here.…”

“…Previous work [24] made the strong assumption that all objects in the environment have straight edges aligned along Cartesian axes, exploiting the fact that many man-made environments are based on square grids. Strong hierarchical object priors were used by [11] to constrain the interpretation of contacts as known 3D object forms. Here we use a prior whose strength lies somewhere between these strong prior approaches and the weak prior blob method of sec.…”

Towards hierarchical blackboard mapping on a whiskered robot

“…Individual components of such a system have previously been investigated in isolation, including whiskered texture recognition [15], [26], [11], [17], [32], surface shape recognition [30], [24], [21], [13], and object recognition [19]. These components have previously been tested under ideal laboratory conditions or in individual mobile settings [41]; here we present steps integrating them into a single platform for hierarchical object recognition, along with results and observations on their performance 'in the wild' in a common arena environment.…”