Scribd
Upload
7 views17 pages

Roscon2019 Slamtoolbox

The document provides an overview of the SLAM Toolbox, detailing its capabilities in mapping and localization within dynamic environments. It discusses the components of graph-based SLAM, various mapping and localization modes, and features such as lifelong mapping and toolbox processing utilities. Additionally, it outlines future developments and ongoing work related to multi-robot systems and enhanced mapping techniques.

Uploaded by

nihokex389
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
7 views17 pages

Roscon2019 Slamtoolbox

The document provides an overview of the SLAM Toolbox, detailing its capabilities in mapping and localization within dynamic environments. It discusses the components of graph-based SLAM, various mapping and localization modes, and features such as lifelong mapping and toolbox processing utilities. Additionally, it outlines future developments and ongoing work related to multi-robot systems and enhanced mapping techniques.

Uploaded by

nihokex389
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

On Use of the SLAM Toolbox:

A Fresh(er) look at Mapping and Localization for the Dynamic World

Steve Macenski, Samsung Research America


Confidential - Do Not Distribute - Simbe Robotics, Inc.
OVERVIEW

Background and Motivating Problems

SLAM Toolbox

Use, Configuration, and Examples

In the Wild and Roadmap


Background - What is SLAM? How does it look like to ROS (REP 105)?

SLAM: Simultaneous Localization and Mapping /map

Receive sensing from the environment


- Range Finders /odom
- Odometry Sources (encoders, IMU, etc) Positioning
- Camera
- Radar

Create a model of the world for positioning


- Sometimes used navigate Fused Odometry
- May be visualized as grids or feature maps

An REP-105 compliant implementation will


provide the map -> odom transform, but often more.
Background - A 30,000 ft View*

GMapping (pf based)


- Presented in 2007 - year of the first iPhone
- Mapping large spaces is an “art”
- In my experience, diverges in >20,000 sqft

Karto (graph based)


- Excellent scan matcher
- Basis of several companies SLAM
- In my experience, an acceptable starting point

Cartographer (graph based)


- Localization, submap serialization, 3D support
- Discussions with Google haven’t inspired confidence
- In my experience, couldn’t reliably create useable maps
● Thick hairy walls, missing simple loop closures, etc

* Look, I can’t talk about everything, we’ll be here all day!


Background - Quick Primer on Graph-based SLAM

4 Major Elements of Graph-based SLAM


- Sensor Matcher
- Pose-Graph
Δ, Σ
- Loop Closure Candidate Selector
- Graph Optimizer
Δ, Σ
Sensor matcher computes relative poses and
confidences between candidate scans

Those are added as edges to a pose-graph

Loop closure candidates identified


and correlated for new constraints
Start
Optimizer tries to reduce error in the graph
Background - Quick Primer on Graph-based SLAM

4 Major Elements of Graph-based SLAM


- Sensor Matcher
- Pose-Graph
Δ, Σ
- Loop Closure Candidate Selector
- Graph Optimizer
Δ, Σ
Sensor matcher computes relative poses and
confidences between candidate scans

Those are added as edges to a pose-graph

Loop closure candidates identified


and correlated for new constraints
Start
Optimizer tries to reduce error in the graph
SLAM Toolbox - An Overview

Map Large and Dynamic Spaces

Reliably Map Spaces >>100,000 sqft Quickly

Improved Ease of Mapping

Rich Set of Toolbox Utilities

Localization that Embraces Change

Online and Offline Processing

Continue Mapping from Prior Sessions Losslessly


Roughly 1 hour to map 45,000 sqft
Courtesy: SimbeRobotics.com
SLAM Toolbox - Mapping

Mapping Modes
● Synchronous: All measurement processing
○ May lag in large spaces if compute is limited
○ Offline: for highest quality
● Asynchronous: Best-effort processing
○ Online: navigating on limited compute

Ceres Plugin Optimizer

Serialization for Continued Mapping / Post-Processing

Map ~200,000 sqft Synchronously in Real-Time on Robot


● Much larger available in asynchronous & cloud processing

[Soon] Multi-robot distributed mapping


SLAM Toolbox - Localization

Takes Advantage of Data + Graph


- Goal: Localization as close to SLAM as possible
- Localize with map and current state in local horizon
- New obstacles become features, not deviations!
- If no base map given, “local SLAM”

Elastic Pose-Graph Deformation


1. Fixed size buffer of new scans
2. Add new node/constraints to graph
3. Once expired from buffer
● remove & free data / nodes
4. Pose-graph reverts to original state

Buffer size set by scan_buffer_size (10)

t
SLAM Toolbox - Continue/LifeLong Mapping

Continue Mapping = Refine or explore new spaces


Each mode can continue building a map
- Load graph, relocalize to a pose or keynode
- Match against local environment for a refined fit
- Find transform from current & serialized frames
- Continue refining map

Lifelong Mapping = Above + removal of extra nodes


* Cartographer’s Lifelong mapping is my continue mapping, it does not bound computation

[Experimental] A living, breathing map


- Online updates to map over many sessions
- Online removal of outdated/redundant data
- Currently computed by a (questionable) heuristic
SLAM Toolbox - Toolbox Processing

I have pose-graphs and scans serialized, what can I do with it?

Batteries Included:
● Kinematic map merging
● Manual pose-graph manipulation
● Rotate maps without aliasing pixels
● Continue to build maps / localization

Ideas:
● Map rendering with different parameters
● Merge N maps using graph structure / data
● Select & remove nodes/constraints
● Annotations
● Analyze odometric & sensor poses to model odometry drift
Configuration and Use

Select Parameters - See documentation for full list and descriptions

stack_size_to_use: Stack size, for serialization/deserialization only

tf_buffer_duration: TF buffer storage duration (offline speedups)

solver_plugin: Optimizer plugin to use

mode: Mapping or localization mode for optimizer settings

map_file_name: Base filepath of serialized map to load


map_start_{pose, at_dock}: Whether to start at a specific pose or dock

enable_interactive_mode: Enables interactive pose-graph manipulation


Examples

Cars!

Sidewalk of San Francisco


What’s Next?

The Work is Never Done

#76 - Multi-laser Robots

#68 - Multi-robot Cloud Processing Sample

Additional Tools for Processing Pose-Graph/Data


● Map merging method using graph structure
● Robot odometry drift estimation
● Selectively remove nodes/constraints

Promote “true” Lifelong Mapping Node to Production


In The Wild
Navigation2 Tiny Ticket Competition Winner - Shivang Patel
Repository, Documentation, and Issue Tracker:
https://github.com/SteveMacenski/slam_toolbox.git

ROS Wiki Page: http://wiki.ros.org/slam_toolbox

Thanks to Other Active Contributor: Ivona Jambrečić

Thanks to Circuit Launch & Andrew Etherington for their


help in finishing this work with facilities & a LIDAR to test

Steve Macenski
Open Source Robotics Engineering Lead
[email protected]
[email protected]

Confidential - Do Not Distribute - Prepared for SoftBank Robotics

You might also like