Learn The Key Components Of HD Map Building!



Learn The Process
From Raw Data Collection
To Consistent, Accurate, Georeferenced Map

Integrate data from multiple trials


Reduce GNSS error


Use affordable hardware


Fuse separate trajectories from multiple sources
such as GNSS, Visual Odometry and Lidar Odometry


Build your own
Mobile Mapping System


Learn how to index spatial data
and obtain rapid access
to high volume data sources

15+ Years Of Experience In SLAM


Building and working with multiple sensor setups

from robotic applications

through stationary Terrestrial Laser Scanners

to the Mobile Mapping Platforms







Your instructors

Janusz Będkowski, PhD, DSc​

Janusz Będkowski is D.Sc. and Ph.D. in mobile robotics. He is a researcher at Polish Academy of Science, Institute of Fundamental Technological Research, and former engineer in TomTom International BV. He is actively working on the theoretical and practical aspects of simultaneous localization and mapping applications assuming global scale. Recently, he is working on the methodology of fulfilling the gap between geodesy and cartography, geo science and mobile robotics. He is active member of European Land Robotic Trial (ELROB) and the European Robotics Hackathon (ENRICH).

Co-funder of HDMapping.org

Karol Majek, PhD

Karol Majek, Ph.D. in Automation and Robotics. Since 2011 he has successfully participated in several mobile robotics competitions including DARPA VRC, ELROB, Eurathlon, Udacity Challenge, Self-Racing Cars, F1/10. 2013-2015 he was involved in the Institute of Mathematical Machines in robotic research projects funded by the EU. In 2016-2017 he was a mentor in the Self-Driving Car Nanodegree by Udacity. In 2018 he received a research grant in the TensorFlow Research Cloud program and in 2019 he defended Ph.D. thesis at Poznań University of Technology: “Automatic selection of deep neural network parameters in mobile robotics”. 2018-2020 he was working on accelerating perception of inspection robots using deep neural networks for object detection at Research and Academic Computer Network – National Research Institute (NASK PIB). He is focused on solving vision problems with deep neural networks. His research interests are: object detection, semantic segmentation, panoptic segmentation, deep neural network inference on low power devices.

Co-funder of HDMapping.org













HD Mapping Program contents

What you will get if you join now:




8 module program with bonuses




Sample dataset from 2 cities - 200+km total
Two Lidar configurations:
Livox Horizon/Avia


Knowledge how to build
an affordable Mobile Mapping System


Knowledge how to reduce errors from pure GNSS
Obtain the accuracy even without INS


Software for building HD Maps
without INS or RTK using affordable hardware

Not a yet-another-framework nor dependency!
Pure Eigen-based optimization


You will learn the math on how to build
optimization systems for large scale mapping


You will know how to reduce overall cost
of your mobile mapping system





Module 1
Introduction

  1. Welcome
  2. How to work with this program
  3. Process of building HD Map
  4. Install required software on Windows
  5. Install required software on Linux
  6. Download datasets

Module 2 Building Mobile Mapping System

  1. Mobile Mapping Systems – SotA
  2. Data sources
  3. Sensor fusion
  4. Data synchronization
  5. Working with ROS
  6. How to build a low cost mobile mapping system
  7. Common problems

Module 3
Data Preprocessing


  1. Lidar odometry
  2. Point cloud indexing
  3. Point cloud decimation
  4. Data filtering
  5. Monocular Visual odometry
  6. Camera calibration
  7. Common problems

Module 4
Point Cloud Registration



  1. 100 variants of the multi view point cloud data registration methods
  2. Point to point
  3. Normal Distributions Transform
  4. Point to projection onto plane
  5. Point to plane using dot product
  6. Distance point to plane
  7. Plane to plane
  8. Pose Graph SLAM
  9. PCL
  10. GTSAM
  11. Manif

Module 5
Semi Rigid Registration


  1. Point cloud data indexing
  2. Point cloud data decimation
  3. Data filtering
  4. Processing of multiple trajectories
  5. Pose Graph SLAM
  6. Multi-view NDT
  7. Generalized ICP
  8. Robust cost functions
  9. Pose interpolation

Module 6 Georeferencing​


  1. Common problems
  2. Georeference data sources
  3. Georeferencing to ground truth point cloud data
  4. Georeferencing to ground truth positions
  5. Georeferencing to ground truth poses
  6. Georeferencing to other geometric primitives
  7. Uncertainty measurement

Module 7
Image Semantic Segmentation


  1. Semantic segmentation problem statement
  2. Segmentation – State of the Art
  3. Frameworks
  4. Road segmentation datasets
  5. Training semantic segmentation network in PyTorch
  6. Delpoyment of trained semantic network

Module 8
Localization


  1. Localization in the map
  2. Key components in localization
  3. Particle Filter
  4. Online data preprocesing
  5. Particle Filter using GPU

BONUS #1

Comparison on available open source datasets


Improve ground truth estimations

Using what you learned in the program!


We will guide you how to do it on publicly available datasets




Office

5 scans

Facade

8 scans

Courtyard

8 scans

Arch

5 scans

Trees

6 scans


12 scans



6 scans



5 scans



3 scans



26 scnas


WHU-TLS Subway station

IMAGER 5010C - 6 scans

WHU-TLS Railway


VZ-400 - 8 scans

WHU-TLS Mountain


ScanStation C5 - 6 scans

WHU-TLS Park

VZ-400 - 32 scans

WHU-TLS Campus

VZ-400 - 10 scans

WHU-TLS Residence

Leica P40 - 7 scans

WHU-TLS River bank

VZ-400 - 13 scans

WHU-TLS Heritage building

VZ-400 - 9 scans

WHU-TLS Excavation

VZ-400 - 12 scans

WHU-TLS Tunnel

VZ-400 - 7 scans

BONUS #2


HDMapping dataset using low cost hardware


Sample dataset from 2 cities - 200+km total

Two Lidar configurations:
Livox Horizon/Avia

Affordable hardware
GNSS, Lidar, Camera


2 cities


200+ km trajectory



Using affordable hardware



Before optimization


Fully processed


Program closed

100 Variants Of The Multi View Point Cloud Data Registration Methods


Enroll Until

Wednesday,

November 30, 2022

[now closed]





Is There Any Other Way?


Yes, there is!

You can waste a lot of resources while experimenting with many available methods. There is no warranty you will achieve your goals in finite time. You can start with many available open source frameworks, e.g. G2O, GTSAM, Ceres and go through all the problems alone.

You will need to answer how to:
  • compensate time offset?
  • process 50km trajectory at once?
  • fix the stale GNSS after motion stopped?
  • merge multiple trajectories into consistent, accurate map?
  • georeference the data?

Frequently Asked Questions

Do I receive all the materials at once, just after paying for them?

NO, this is a PRESALE. After PRESALE ends, we will focus on preparing video tutorials, which we will release in batches in 2023 Q1. The code and methods are ready.

When will the Course be Released?

We plan to send first materials in 2023 Q1. All the necessary software is already prepared and published on GitHub

Will I have lifetime access to this course?

YES, access to this program is not limited. This is a self-paced course. You can take this course at any time. 

Is there a Refund Policy?

YES, 30-day money-back satisfaction guarantee. Counting starts when you receive first materials in 2023. If you are not satisfied, just write an email to [email protected], we do not ask for the reason.

Will I receive a certificate for this training?

NO, we do not offer certificates.

Can I work on Windows/Linux?

YES, both Windows and Linux are supported

Do I need GPU enabled PC?

GPU is optional - will be used only in modules 7 and 8.

Can I optimize large trajectories on a laptop/desktop?

You will need a decent laptop or desktop - it should be enough. We suggest to have at least 16/32 GB RAM.

Do I need a datacenter to start?

NO, you can start with a single laptop/desktop PC with 1-2TB storage.

Can I join in December 2022 or in January 2023?

NO, we will close this presale and start to focus on providing the best materials we have. You will need to wait at least until June 2023.

Our team is focused on other projects, can we buy it later?

Only in this PRESALE or later in 2023. We guarantee the lowest price now. Your team can join now and wait until they have the capacity. This is a self-paced, online course with lifetime access.

Why not use frameworks such as G2O, GTSAM, Ceres?

It is hard to integrate them in practical applications -introduction of specific, new observations types.

We have complex procedures, and will not make it till December, what we can do?

Please contact us and place an order by e-mail [email protected]

Other questions?

Please contact us via email [email protected]

What you will get if you join in the PRESALE?


8 module program with bonuses
Online, self-paced course with lifetime access
100 Variants Of The Multi View Point Cloud Data Registration Methods
Sample dataset from 2 cities - 200+km total
Knowledge how to build an affordable mapping system
Knowledge how to reduce errors from pure GNSS, without RTK
You will learn the math on how to build optimization systems for large scale mapping
You will know how to reduce overall cost of your mobile mapping system
Software for building HD Maps without INS or RTK using affordable hardware
30-day money-back satisfaction guarantee
Best price guarantee - we will not reduce the price, you get the best offer now



How this PRESALE works?

We gather the first batch of minimum 20 people
Money back guarantee if the goal is not reached
We work to deliver the best materials
We deliver first modules in 2023Q1
We will not reopen this program before June 2023

Program is closed since


Wednesday,

November 30, 2022