Environment Maps are consumed and used by various components in the Isaac stack. This guide takes you through the sequential process of creating maps to enable autonomous navigation.
The process of map creation is described in the following workflow diagram:
You need the following before you create maps:
Nova Carter Robot with 2TB of available space.
PC (with browser) connected to the same WiFi Network as the robot.
Joystick controller, connected to the robot.
Layout or detailed schematic of the facility to be mapped (as a PDF). The layout must be highlighted to specify where the stairs and other “No-Go” zones are (for semantic labeling).
Ethernet connectivity with an RJ45 connector (for fast uploading of data).
Internet access.
Sensor calibration file for the specific robot being used. Refer to the Sensor Calibration section for details.
Any specific guidelines you have received from the NVIDIA Solutions team (for example, the recommended data collection route).
To generate a map, data must be gathered from the Nova Carter’s stereo camera and Lidar sensors. Nova Carter is equipped with an application designed to streamline this data collection process. This application is responsible for capturing information from the front HAWK stereo camera as well as the 360-degree Lidar sensor. Subsequently, you have the option to upload this data to a server to initiate map creation. The entire data collection and upload procedure is guided by a remote user interface.
Typically, constructing a map for a facility spanning 100,000 square meters necessitates approximately 500 gigabytes (GB) of data.
Workflow Overview
The following workflow diagram describes the data collection process for map creation:
Data Recording and Upload
Refer to Record and Upload Tool for details on how to launch the recording app. It also provides instructions for collecting and uploading data.
We advise conducting a brief data validation process prior to commencing full-scale data collection. While this step can require up to one full workday to complete, it has the potential to significantly reduce overall time.
Data Validation (Recommended)
Beyond assessing data suitability for map creation, data validation also provides confirmation of the verified connection between the robot and cloud storage, ensuring the data’s usability in the mapping process.
Follow these rules for quick data collection:
Launch the Record and Upload Tool.
Press the START utton to start the Data Recorder application.
Use the joystick controller to move the robot 10m straight ahead at speeds less than miles/hour (<1.3m/s). Nova Carter is limited to a speed of 1.1m/s, so anything less than maximum speed is sufficient.
NoteThe operator should not move during this period.
Stop, then continue moving again for another 2-3 seconds.
Press the STOP button to stop the Recorder App and upload the data for validation.
Share the name of the .pod file you uploaded with the NVIDIA Solutions team.
The NVIDIA Solutions team returns data validity approval to you, typically within one work day.
Full Data Collection
After receiving confirmation of data validity, you can proceed with complete data collection.
Data Collection Guidelines
Here are some guidelines to follow while collecting data:
Opt for data collection during periods with minimal human presence.
If feasible, lower window blinds to reduce unwanted reflections.
Maintain a minimum distance of 2 meters from the robot as an operator to prevent sensor obstruction (preferably around 10 meters).
As the operator, stay behind the robot as it moves.
Ensure that the Nova Carter maintains a minimum distance of 10 centimeters from walls and other surfaces.
Aim to cover the majority of the space by guiding the Nova Carter to each corner.
When collecting data in open areas, plan routes that are spaced approximately 5 meters apart (refer to the provided image).
In open areas, ensure loop closure by revisiting the same locations during data collection.
Aim to encompass all areas and complete the data collection in a single session. If multiple sessions are required, endeavor to commence each new session from the point where the previous one concluded.
Additional Guidelines
Use Chrome if possible.
Record in an environment with good Wi-Fi coverage. Operating in areas with poor Wi-Fi can cause the user interface to become unresponsive. Refresh the browser if this occurs.
Connect to Ethernet before uploading data.
Data Collection Steps
Start the Data Recorder application.
Press the START Button to start the Data Recorder application.
Move the robot as per guidelines above.
Press the STOP button to stop the Recorder App and upload the data for validation.
Share the name of the uploaded .pod file to the NV Solution Engineer.
Quit the application from the PC once done.
Triggering Map Creation
After all data has been uploaded, send an Email to the NVIDIA Solutions team with the following information:
The project name (for example, the name of the building).
The approximate latitude and longitude where the data collection occurred.
A Google Maps or OpenStreetMap screenshot indicating the approximate boundary of the data collection.
You receive a notification after the occupancy map has been created. Typically, the occupancy map is ready in about two work days after data collection is uploaded.
You can label the semantic map using the Map Annotation Tool. Alternatively, if it’s part of your contract, the NVIDIA Solutions team can provide the semantic map; the delivery time depends on the map size and can take one or two extra work days.
Downloading and Accessing the Map
The Mapping service provides three types of map:
2D occupancy map
Semantic map
3D occupancy map
If you have access to the Map Annotation Tool, you can use it to download the 2D occupancy map and semantic map. If not, you can request the NVIDIA Solutions team to provide the map files.
If you are following the Running the Navigation Stack with Isaac AMR Cloud Services workflow, and choose to use maps served by DeepMap, you must find the Map UUID in the Annotation Tool and provide it as part of the Navigation Stack config. It then automatically downloads the latest released 2D occupancy map and semantic map. For more details, see Running the Navigation Stack with Isaac AMR Cloud Services.
The output map (2D Occupancy map) is a result of the Lidar field of view (FOV). Therefore, stairs that descend, or other artifacts below the floor might not be included in the map if they were not observed by the robot lidar; these must be added in the semantic map.
The 3D occupancy map is provided by the NVIDIA Solutions team in the form of a .ply file.
To sign up for the Map Annotation tool, contact the NVIDIA Isaac Mapping team and provide the following information:
Email address
User name
Name of the company or organization
The NVIDIA Solutions team creates an account for the user, usually within 24 hours. You then receive an email from no-reply-deepmap@nvidia.com with the initial password and user name. You can log in to the tool hosted at https://robotictools.deepmap.com/labeling/login.
The first time you log in, you are required to change the password.
Follow the steps in the Map Annotation Tool section to annotate the map.