Analysis of ROS-based Visual and Lidar Odometry for a Teleoperated

Crawler-type Robot in Indoor Environment

Maxim Sokolov, Oleg Bulichev and Ilya Afanasyev

Institute of Robotics, Innopolis University, Universitetskaya str. 1, 420500 Innopolis, Russia

Keywords:

Monocular SLAM, ROS, Visual Odometry, Lidar Odometry, Crawler Robot, ORB-SLAM, LSD-SLAM.

Abstract:

This article presents a comparative analysis of ROS-based monocular visual odometry, lidar odometry and

ground truth-related path estimation for a crawler-type robot in indoor environment. We tested these methods

with the crawler robot ”Engineer”, which was teleoperated in a small-sized indoor workspace with ofﬁce-

style environment. Since robot’s onboard computer can not work simultaneously with ROS packages of lidar

odometry and visual SLAM, we used online computation of lidar odometry, while video data from onboard

camera was processed ofﬂine by ORB-SLAM and LSD-SLAM algorithms. As far as crawler robot motion is

accompanied by signiﬁcant vibrations, we faced some problems with these visual SLAM, which resulted in

decreasing accuracy of robot trajectory evaluation or even fails in visual odometry, in spite of using a video

stabilization ﬁlter. The comparative analysis shown that lidar odometry is close to the ground truth, whereas

visual odometry can demonstrate signiﬁcant trajectory deviations.

1 INTRODUCTION

Over the last decade visual odometry has become the

valuable tool for estimation of a vehicle’s pose, orien-

tation and trajectory through analysis of correspond-

ing onboard camera images recorded during vehicle

motion. However, visual odometry methods are sen-

sitive to illumination conditions and can fail in case of

insufﬁciency of visual features since a scene requires

enough texture to let explicit motion be estimated.

It makes essential to combine visual odometry with

other measurements like wheel odometry, lidar odom-

etry, global positioning system (GPS), inertial mea-

surement units (IMUs), etc. (Scaramuzza and Fraun-

dorfer, 2011), (Zhang and Singh, 2015), (Sarvrood

et al., 2016). Nevertheless, the other types of on-

board odometry can have own drawbacks. For in-

stance, lidar odometry at robot motion includes ﬂuc-

tuations in positions of point clouds over time. There-

fore, onboard sensor odometry should be provided

with ground truth

Since we utilize a single onboard camera, we

are interested in monocular visual odometry, which

is characterized by computing both relative vehi-

cle motion and 3D scene structure from 2D video

”Ground truth” is deﬁned as a reference tool or a set

of measurements that are known to be much more accurate

than measurements from a system under investigation

data. Very often, researchers deﬁne three cate-

gories of monocular visual odometry: (1) feature-

based, (2) appearance-based (also known as direct),

and hybrid methods (Scaramuzza and Fraundorfer,

2011). Appearance-based methods estimate intensity

of all image pixels with the following direct image-

to-image alignment, whereas feature-based methods

extract appreciable and repeatable features that can

be tracked through the frames. Finally, hybrid meth-

ods apply to a combination of both previous ap-

proaches. We need to emphasize that in our investiga-

tion we exploit visual simultaneous localization and

mapping (V-SLAM) methods, which are originally

focused on calculating a global map and robot path

while tracking onboard camera position and orienta-

tion. However, in our case we study a robot motion

in small indoor workspace, thereby we extend the re-

sults of V-SLAM to visual odometry, neglecting a dif-

ference between these deﬁnitions and considering the

workspace map as a global map. In our research we

use both feature-based and appearance-based meth-

ods for robot path recovery by choosing two monoc-

ular V-SLAM: ORB-SLAM (Mur-Artal et al., 2015)

and LSD-SLAM (Engel et al., 2014) recognized as

enhanced and robust methods. Both of these meth-

ods have demonstrated good results in tracking, map-

ping, and camera localization for outdoor applica-

tions, but there are some uncertainties with robust-

316

Sokolov, M., Bulichev, O. and Afanasyev, I.

Analysis of ROS-based Visual and Lidar Odometry for a Teleoperated Crawler-type Robot in Indoor Environment.

DOI: 10.5220/0006420603160321

In Proceedings of the 14th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2017) - Volume 2, pages 316-321

ISBN: Not Available

ness and feasibility to indoor robot navigation in typ-

ical ofﬁce-style environment with monotone-painted

walls (Buyval et al., 2017). Except V-SLAM we also

provided ROS-based onboard lidar odometry in on-

line mode, getting more accurate information about

robot motion trajectory. Since crawler robot motion

suffers signiﬁcant shakes, vibrations and sharp turns,

it brings deviations in onboard sensor outputs, de-

creasing accuracy of trajectory evaluation or even fail-

ing visual odometry. Therefore, robot path veriﬁca-

tion is provided by an external camera-based moni-

toring system, which is not affected by vibrations.

Our main goal of this study is to compute a

crawler-type robot trajectory via different odome-

try methods realized in ROS

, providing: (1) vi-

sual odometry acquisition for robot motion within

the workspace by using onboard camera and two V-

SLAM packages: feature-based ORB-SLAM and di-

rect LSD-SLAM; (2) comparison of onboard visual

and lidar odometry; (3) a veriﬁcation of both visual

and lidar odometry with the real robot trajectory mea-

sured by the external camera-based monitoring sys-

tem as the ground truth. To summarize, this pa-

per analyzes and compares different monocular visual

odometry with lidar odometry and the ground truth-

based robot path estimation. Since onboard computer

of our robot did not permit a simultaneous run of ROS

packages for lidar and two visual SLAM, we launched

ROS-based lidar odometry online and recorded on-

board and external cameras’ video in synchronous

mode. Then, we processed onboard video ofﬂine with

ORB and LSD-SLAM, comparing visual and lidar

odometry with ground truth-based robot trajectories.

The rest of the paper is organized as following.

Section 2 introduces system setup, Section 3 presents

ROS-based visual and lidar odometry, and Section 4

describes indoor tests and robot trajectories evalua-

tion. In Section 5 we analyze visual and lidar odome-

try estimation. Finally, we conclude in Section 6.

2 SYSTEM SETUP

2.1 Robot System Conﬁguration

The crawler-type robot ”Engineer” (Fig. 1) is de-

signed and manufactured by ”Servosila” company

overcome complex terrain while navigating in con-

ﬁned spaces and solving a number of challenging

tasks for search and rescue missions. The robot has

Robot Operating System (ROS), which is a set of soft-

ware libraries and tools for robot applications, www.ros.org

”Servosila” company, www.servosila.com/en/

tracks, ﬂippers, and a robotic arm, that support capa-

bilities for climbing stairs, traversing doorways and

narrow passages, negotiating obstacles, leveling itself

from sideways/upside down positions, etc. Moreover,

the robotic arm is equipped by a head with sensors,

controllers and grippers, that allows to have a ﬂexible

remote control for lifting heavy loads, opening differ-

ent doors, plus grasping, pushing or pulling objects.

The robot sensors pack can contain of a laser scan-

ner, an optical zoom camera, a thermal vision cam-

era, a pair of stereo cameras, IMU and GPS naviga-

tion tools. The detailed information about our set of

robot vision system is presented in Table 1.

Since our robot’ sensor pack does not have a

built-in laser scanner, we mounted the scanning laser

rangeﬁnder Hokuyo URG-04LX-UG01

on the robot

head (see, Fig. 2). This laser is often used for au-

tonomous robots, as far as it has light weight (160g),

low-power consumption (2.5W), convenient for in-

door application scanning range from 2 cm to 5.6 m,

wide ﬁeld of view up to 240

◦

, angular resolution of

0.36

◦

, refresh rate of 10Hz, and accuracy up to 3%

relatively to measured distance.

Figure 1: The Servosila ”Engineer” crawler robot. Courtesy

of Servosila company.

Figure 2: The ”Engineer” robot head’s sensor system.

2.2 Robot Vision System

To validate onboard visual and lidar data, we used ex-

ternal ground truth system based on Basler acA2000-

Hokuyo Automatic Co. 2D lidar, www.hokuyo-aut.jp

Analysis of ROS-based Visual and Lidar Odometry for a Teleoperated Crawler-type Robot in Indoor Environment

317

Table 1: The Servosila ”Engineer” robot vision system.

Type Quantity Direction Resolution Palette Night vision Zoom Rate

Mono one front 1280*720 RGB No Yes 60fps

Stereo pair two front 640*480 YUYV Yes No 30fps

Mono one rear 640*480 YUYV No No 30fps

Figure 3: Workspace with the ground truth camera above.

50gc camera

(see, characteristics in Table 2), which

was hung up under the workspace on the height of

about 3m and connected to PC (Fig. 3). Before tests

the camera was calibrated against a chessboard with

”camera calibration” ROS package

Table 2: Basler acA2000-50gc camera characteristics .

Parameter Conﬁguration

Image Sensor CMV2000 CMOS

Video resolution 2MP, 2046*1086

Frame rate 50 fps

Mono/Color Color

Shutter Global shutter

3 ROS-BASED VISUAL AND

LIDAR ODOMETRY

3.1 ROS-based ORB SLAM Odometry

Oriented FAST (Rosten and Drummond, 2006) and

Rotated BRIEF (Calonder et al., 2010) algorithm,

Area Scan Camera from Basler AG company,

www.baslerweb.com/en/products/cameras/

”camera calibration” ROS package is based on

OpenCV library: wiki.ros.org/camera calibration

which form ORB SLAM

method is a feature-based

real-time SLAM library for monocular, stereo and

RGB-D cameras (Mur-Artal et al., 2015). It is able

to build a sparse 3D scene and to compute an on-

board camera trajectory, thereby recovering a vision-

based robot odometry. In cases of heterogeneous

environment, ORB-SLAM demonstrates robustness

to complex motion clutter, performing wide baseline

loop detection and real time automatic camera re-

localization. To process live monocular streams, a li-

brary with a ROS node is used.

The ORB-SLAM adapts main ideas of previous

SLAM works, such as Parallel Tracking and Map-

ping (PTAM, (Klein and Murray, 2007)) and Scale

Drift-Aware Large Scale Monocular SLAM (Strasdat

et al., 2010). Thus, it uses advanced approaches to lo-

calization, loop closing, bundle adjustment, keyframe

selection, feature matching, point triangulation, cam-

era localization for every frame, and relocalization

after tracking failure. Moreover, ORB-SLAM sur-

passes PTAM algorithm, providing camera tracking

with ORB-features extraction (Rublee et al., 2011),

scale-aware loop closing, co-visibility information for

large scale operation, occlusion handling, and invari-

ance to viewpoint at relocalization (Mur-Artal et al.,

2015). ORB-SLAM key properties include:

• the same features for tracking, mapping, re-

localization and loop closing;

• real time loop closing, which is based on the opti-

mization of a pose graph;

• invariant relocalization of real time camera to

viewpoint and illumination.

ORB-SLAM is a real-time SLAM library in

ROS, which provides both necessary calculations and

graphical user interface (GUI) for visualization of a

camera trajectory, a sparse 3D reconstruction, and

real-time features on video frames. To launch ORB-

SLAM algorithm in ROS, the following operations

should be executed:

• building a node for Monocular SLAM;

• running Monocular Node:

rosrun ORB SLAM2 Mono path to vocabulary

path to webcam settings ﬁle

ORB-SLAM2 package is available at

https://github.com/raulmur/ORB SLAM2

ICINCO 2017 - 14th International Conference on Informatics in Control, Automation and Robotics

318

Monocular SLAM node maintains three parallel

threads: Tracking (to help in a camera localization),

Local Mapping (to build a new map) and Loop Clos-

ing (to obtain a closed-loop trajectory).

3.2 ROS-based LSD-SLAM Odometry

Large-Scale Direct Monocular SLAM

(LSD-SLAM)

creates a real-time global, semi-dense map in a fully

direct mode without using keypoints, corners or any

other local features (in opposite to feature-based

methods like ORB-SLAM). Direct featureless track-

ing is performed by image-to-image alignment us-

ing a coarse-to-ﬁne approach with a robust Huber

weights (Engel et al., 2014). Depth is estimated only

using pixels near image boundaries and semi-dense

maps (which are denser than maps of feature-based

methods) are created continuously. LSD-SLAM

forms both a camera trajectory and a semi-dense 3D

scene reconstruction, where a global mapping is built

by performing a pose graph optimization. The ad-

vantage of LSD-SLAM over feature-based methods

is in reconstruction of a more complete 3D scene with

textured smooth surfaces, which could be missed by

feature-based algorithms.

LSD-SLAM is a fully direct method, which is ca-

pable to build real-time large-scale semi-dense maps

using an onboard computer. LSD-SLAM is launched

in ROS with:

• Starting the camera driver:

roslaunch webcam camera usb camera.launch

• Launching the LSD-SLAM viewer:

rosrun lsd slam viewer viewer

• Initiation of the main node:

rosrun lsd slam core live slam

/image:=/usb cam node/image raw

/camera info:=/usb cam node/camera info

Note that LSD-SLAM consists of two ROS pack-

ages: (1) lsd slam core that includes full SLAM sys-

tem, which provides live camera operation live slam

using ROS input/output, and (2) lsd slam viewer,

which is optionally used for 3D visualization.

3.3 ROS-based Lidar Odometry

To obtain ROS-based lidar odometry for ”Engineer”

robot motion within indoor workspace, we used the

open source hector slam package

, which contains of

modules related to SLAM execution in unstructured

LSD-SLAM package is available at

https://github.com/tum-vision/lsd slam

ROS package hector slam is available at

https://github.com/tu-darmstadt-ros-pkg/hector slam

Figure 4: The block-scheme of the ”Engineer” robot trajec-

tory evaluation with onboard sensors and ground truth.

environments (Kohlbrecher et al., 2013). Its algo-

rithm relies on fast lidar data scanning and matching

at full lidar refresh rate. To stabilize the laser scan-

ner data, hector slam combines with an attitude esti-

mation and an optional pitch and roll angles, calculat-

ing probable locations and environment maps even for

rugged ground terrain (Kohlbrecher et al., 2013). To

recover a trajectory of moving robot with lidar data,

the ROS module hector

tra jectory server

keeps

tracking of multiple coordinate frames over time.

3.4 Indoor Tests with Crawler-type

Robot

Our goals in these indoor tests with the crawler robot

are: (1) to calculate visual odometry from the robot

motion within the workspace by using two ROS-

based V-SLAM packages: ORB and LSD-SLAM; (2)

to obtain onboard lidar odometry; (3) to verify the vi-

sual and lidar odometry with a trajectory calculated

by video processing of the external ground truth cam-

era. The block-scheme of the ”Engineer” robot path

evaluation with onboard sensors and ground truth is

shown in Fig. 4

At the initial stage of our experiments we put the

external ground truth camera above the workspace,

and calibrated both onboard and external cameras

with a chessboard at different locations. Our tests

were performed with a human-operated crawler-type

robot ”Engineer” followed a close-loop trajectory in

a small-size indoor workspace with ofﬁce-style en-

vironment and partially glass walls (Fig. 5a). The

data was recorded with onboard sensors: 2D lidar

and high-resolution camera with a global shutter and a

ROS package hector tra jectory server is a part of

hector slam package, wiki.ros.org/hector trajectory server

Analysis of ROS-based Visual and Lidar Odometry for a Teleoperated Crawler-type Robot in Indoor Environment

319

(a) Test 1: complex robot motion.

(b) Test 2: forth-

and-back motion.

Figure 5: Lidar map for different robot trajectories (green

color) with ROS hector slam package in RViz viewer.

frame rate of 60 fps. Since robot’s onboard computer

can not work simultaneously with ROS packages of

lidar odometry and two visual SLAM, we launched

ROS-based hector slam package online and recorded

onboard and external cameras’ video in synchronous

mode. Then, we processed onboard video ofﬂine

with ORB and LSD-SLAM, comparing visual and li-

dar odometry with ground truth-based robot trajecto-

ries. The main drawbacks of the crawler robot mo-

tion are signiﬁcant vibration and sharp turns that typi-

cally lead to poor visual SLAM results or fails of these

ROS packages. To minimize vibration effects on vi-

sual odometry, we provided a series of experiments,

moving robot forth-and-back at slow speed in teleop-

eration mode without turns to either side (Fig. 5b).

However, in spite of our precautions, strong vibra-

tion forced us to stabilize onboard video during post-

processing stage ofﬂine by OpenCV video stabiliza-

tion ﬁlter described in (Grundmann et al., 2011).

4 ANALYSIS OF VISUAL AND

LIDAR ODOMETRY

Lidar odometry. To process Hokuyo 2D Laser Scan-

ner data, we used ROS-based hector slam package,

which calculated 2D map of our indoor environment

and robot trajectory with the following visualization

in RViz

(Fig. 5). However, glass walls nearby the

workspace were transparent for lidar (Fig. 5a) that

can create a problem for autonomous navigation of

the robot in such type of indoor environment.

ORB-SLAM tests. Since ORB-SLAM is feature-

based method, its 3D point cloud is very sparse. How-

ever, ORB-SLAM performs map reconstruction over

selected keyframes with calculation of camera po-

sitions after every recall in all video dataset (see,

Fig. 6), building as well 2D robot motion trajectory.

LSD-SLAM tests. In our experiments with

crawler robot, LSD-SLAM has frequently failed in

conditions of camera vibrations and sharp turns, even

after video stabilization by the OpenCV ﬁlter.

RViz is 3D visualization tool for ROS, wiki.ros.org/rviz

Figure 6: ORB-SLAM map viewer with 3D point cloud

(left), and an image with extracted features (right).

Figure 7: The contrasting white label on the robot’s head

from the ground-truth camera.

Ground truth-based path evaluation. To mon-

itor the robot motion from the height of about 3m,

we used the external high resolution camera Basler

(see, Fig. 3), thus established ground truth. To clearly

identify the robot motion within the workspace, we

placed white label on the robot’s head. Thereby, from

a grayscale video of the ground truth camera we eas-

ily recognized the contrasting white label and robot

contour (see, Fig. 7), computing the robot geometri-

cal center and its corresponding trajectory.

Analysis of trajectories. Finally, we computed

and scaled visual and lidar-based trajectories for the

following comparative analysis. Typical trajecto-

ries are shown in Fig. 8, where green, red and blue

(dot) curves mean robot path evaluation calculated

by monocular ORB-SLAM, ground truth camera and

lidar data correspondingly. We used the root mean

square error (RMSE) as metrics of robot trajectory

evaluation accuracy. The robot path evaluated by ex-

ternal ground truth camera was used as a base path

to rescale all other trajectories in terms of ground

truth axes. Therefore, we developed a software to es-

timate a correspondence between every point of the

base path and the nearest points of another trajec-

tory by computing the square root of mean of squared

distances between these points. Thus, the maximum

RMSE reached 1.96 cm between the ground truth

and lidar-based trajectories, and 12.6 cm between the

ground truth and ORB-SLAM trajectories.

ICINCO 2017 - 14th International Conference on Informatics in Control, Automation and Robotics

320

Figure 8: The robot trajectories computed by monocular

ORB-SLAM (green curve), external ground-truth camera

(red curve), and lidar data (blue dot curve).

5 CONCLUSION AND FUTURE

WORK

In this paper we analyze and compare robot trajec-

tories acquired by onboard 2D lidar and monocular

camera, and evaluated by ROS-based visual odom-

etry, lidar odometry and the external ground truth

camera. We used two visual SLAM methods for the

robot path recovery: feature-based ORB-SLAM and

appearance-based LSD-SLAM, because they usually

demonstrate good results in tracking, mapping, and

camera localization. Our tests were performed with

a human-operated crawler-type robot ”Engineer” fol-

lowed a close-loop trajectory in a small-sized indoor

workspace with ofﬁce-style environment and partially

glass walls. Since onboard computer of our robot can

not work simultaneously with ROS packages of lidar

odometry and two visual SLAM, we used ROS-based

hector slam package online and recorded onboard

and external cameras’ video in synchronous mode.

Then, we processed onboard video ofﬂine with ORB

and LSD-SLAM, comparing visual and lidar odome-

try with ground truth-based robot trajectory. The main

drawbacks of the crawler-type robot motion are sig-

niﬁcant vibration and sharp turns that result in poor

results or even fails in ROS-based visual SLAM pack-

ages. To minimize vibration effects on visual odom-

etry, we moved robot forward and backward at slow

speed in teleoperation mode without sharp turns to ei-

ther side, and used OpenCV video stabilization ﬁl-

ter ofﬂine as post-processing stage. However, in spite

of our precautions the vibrations were so strong that

LSD-SLAM odometry frequently failed and lost the

robot trajectory during its motion.

The comparative analysis of trajectories computed

by ORB-SLAM, lidar and external ground truth cam-

era data allowed to make the following conclusions:

(1) lidar odometry is close to the ground truth path

evaluation; (2) ORB-based visual odometry contin-

ues working in spite of strong camera vibration dur-

ing crawler motion, but its trajectory shows signiﬁ-

cant deviations in comparison with the ground truth.

Our future plans deal with realization of other vi-

sual SLAM and odometry methods, providing tests

in more complex environment and improving experi-

mental technique for better accuracy estimation.

REFERENCES

Buyval, A., Afanasyev, I., and Magid, E. (2017). Compar-

ative analysis of ros-based monocular slam methods

for indoor navigation.

Calonder, M., Lepetit, V., Strecha, C., and Fua, P. (2010).

BRIEF: Binary Robust Independent Elementary Fea-

tures, pages 778–792. Springer Berlin Heidelberg,

Berlin, Heidelberg.

Engel, J., Sch

ops, T., and Cremers, D. (2014). LSD-

SLAM: Large-Scale Direct Monocular SLAM, pages

834–849. Springer International Publishing, Cham.

Grundmann, M., Kwatra, V., and Essa, I. (2011). Auto-

directed video stabilization with robust l1 optimal

camera paths. In Proceedings of the 2011 IEEE Con-

ference on Computer Vision and Pattern Recognition,

CVPR ’11, pages 225–232, Washington, DC, USA.

IEEE Computer Society.

Klein, G. and Murray, D. (2007). Parallel tracking and map-

ping for small ar workspaces. In 2007 6th IEEE and

ACM International Symposium on Mixed and Aug-

mented Reality, pages 225–234.

Kohlbrecher, S., Meyer, J., Graber, T., Petersen, K., Klin-

gauf, U., and von Stryk, O. (2013). Hector open

source modules for autonomous mapping and naviga-

tion with rescue robots. In Robot Soccer World Cup,

pages 624–631. Springer.

Mur-Artal, R., Montiel, J. M. M., and Tard?s, J. D. (2015).

Orb-slam: A versatile and accurate monocular slam

system. IEEE Transactions on Robotics, 31(5):1147–

1163.

Rosten, E. and Drummond, T. (2006). Machine Learn-

ing for High-Speed Corner Detection, pages 430–443.

Springer Berlin Heidelberg, Berlin, Heidelberg.

Rublee, E., Rabaud, V., Konolige, K., and Bradski, G.

(2011). Orb: An efﬁcient alternative to sift or surf. In

2011 International Conference on Computer Vision,

pages 2564–2571.

Sarvrood, Y., Hosseinyalamdary, S., and Gao, Y. (2016).

Visual-lidar odometry aided by reduced imu. ISPRS

International Journal of Geo-Information, 5(1):3.

Scaramuzza, D. and Fraundorfer, F. (2011). Visual odom-

etry [tutorial]. IEEE Robotics Automation Magazine,

18(4):80–92.

Strasdat, H., Montiel, J., and Davison, A. J. (2010). Scale

drift-aware large scale monocular slam. Robotics: Sci-

ence and Systems VI.

Zhang, J. and Singh, S. (2015). Visual-lidar odometry and

mapping: low-drift, robust, and fast. In 2015 IEEE

International Conference on Robotics and Automation

(ICRA), pages 2174–2181.

Analysis of ROS-based Visual and Lidar Odometry for a Teleoperated Crawler-type Robot in Indoor Environment

321