Virtual Occupancy Grid Map with Applications to Autonomous 3D Reconstruction in Underwater Unstructured Scenes - Robotics Institute Carnegie Mellon University

Virtual Occupancy Grid Map with Applications to Autonomous 3D Reconstruction in Underwater Unstructured Scenes

Master's Thesis, Tech. Report, CMU-RI-TR-18-49, Robotics Institute, Carnegie Mellon University, August, 2018

Abstract

Reconstruction of marine structures such as pilings underneath piers presents a plethora of interesting challenges. It is one of those tasks better suited to a robot due to harsh underwater environments. Underwater reconstruction typically involves human operators remotely controlling the robot to predetermined way-points based on some prior knowledge of the location and model of the object of interest. However, it is impractical and dangerous to manually control the robot to perform the reconstruction task in an unstructured scene where prior knowledge of the locations and shapes of objects of interest is not available.

Based on the state-of-the-art mapping and planning methods, this thesis presents an approach that enables the robot to perform reconstruction task in an underwater unstructured scene autonomously without any prior knowledge except the bounding box within which the robot should operate. In particular, the key challenge lies in working with a world representation that is compatible with both mapping and planning algorithms. We address this challenge with the proposed virtual occupancy grid map or VOG-Map. VOG-Map is represented by a collection of local occupancy grid maps whose respective poses are optimized to account for the drift or accumulated noise. Based on the VOG-Map, a path planning algorithm is able to plan safer way-points for collision-free paths as well as more informative way-points for accurate reconstruction than is that based on a global occupancy grid map. By adding additional constraints on the way-points and modifying how their respective information gains are computed, we show how our mapping algorithm could work well with the way-points returned by the planner based on VOG-Map.

The quality of both VOG-Map and scene reconstruction depends on the mapping algorithm. We employ smoothing-based pose graph simultaneous localization and mapping (SLAM) algorithm which is capable of correcting for drift upon loop closures when the algorithm determines that the robot has come back to a previously visited area. However, in scene that lacks geometric structures, the process of determining loop closures via methods such as iterative closest point (ICP) is prone to error. We incorporate the approach of determining degeneracy in the scene into our ICP method and add a loop closure constraint to the SLAM optimization problem, constraining only well-conditioned directions based on principal component analysis. We demonstrate the use of VOG-Map for implementing an underwater system in which the robot actively plans paths to generate accurate 3D scene reconstructions. We evaluate our system qualitatively and quantitatively on simulated as well as real-world experiments.

BibTeX

@mastersthesis{Bing-Jui Ho-2018-107317,
author = {Bing-Jui Ho},
title = {Virtual Occupancy Grid Map with Applications to Autonomous 3D Reconstruction in Underwater Unstructured Scenes},
year = {2018},
month = {August},
school = {Carnegie Mellon University},
address = {Pittsburgh, PA},
number = {CMU-RI-TR-18-49},
keywords = {Mapping, Planning, SLAM, ICP},
}