3D POSITIONING SYSTEM USING A SINGLE CAMERA
Based only on the existing features of the venue, it works for all kind of robots, drones and vehicles
WHAT IS JACK SLAM?
JACK SLAM © is a visual positioning system using only a classic camera to provide an accurate location to all types of robots: AGV, forklifts, drones, service robots, and all automated vehicles.
JACK SLAM © use the Visual SLAM technology, the latest state-of-the-art technology to locate any mobile system indoor.
JACK SLAM ©is dedicated for an indoor usage, but it can be used also outdoor where a better accuracy than GPS is required.
It works in urban and peri-urban environment
JACK SLAM © is a range of hardware and software positioning products to integrate to your mobile robot (AMR/AGV).
HOW TO INTEGRATE JACK SLAM ?
JACK SLAM can be integrated into any system in 3 ways:
– JS-BOX: A ready-to-use box that can be mounted on any mobile element. The JS-BOX includes a permanent user license.
– JS-BOARD: A low-cost board (without case) communicating via its ethernet port. To be integrated into an existing architecture
– As third-party software to be integrated on a customer computer
- Real time: Localize any robot/vehicle with the natural visual features of the venue.
- Low cost: Nothing to install in the environment, works with a standard RGB camera(s).
- Accurate: centimeters localization.
- Reliable: Cope with difficult cases such as light changes, lack of illumination, dynamic objects, texture-less scenes.
- Robust: Data fusion with other sensors (odometry, LIDAR, IMU, …)
- Insensitive to moving objects: humans, animals, other robots, etc.
- Insensitive to modification of the venue if camera is turned toward the ceiling.
- Efficient: Loop closures mechanism generating accurate maps.
- Map update: Automatic each time the AGV returns to an area already visited.
- Map sharing: With all other robots of the fleet.
- Light: Works on any hardware (to get an idea it works at 25fps on a Raspberry pi4)
- Multiple inputs: monocular, stereo and depth cameras
HOW IT WORKS ?
JACK SLAM uses the natural landmarks of the places in which it evolves: light fixtures, corners of doors or windows, …It tracks these landmarks from image to image and locates itself in real time by performing a triangulation.
Depending on the environment, it uses 30 to 300 landmarks in an image to locate.
In the video we see what JACK SLAM is seeing. The green points are the features it takes to perform the localization.
We can see JACK SLAM in action. Its camera is turned at 70° towards the ceiling and 20° forward:
This is an ideal inclination for a localization performed with landmarks on the ceiling and the tops of the walls.
In this video, we also see that direct lights do not affect the localization.
How JACK SLAM works in a real case
Example on an industrial cleaning machine
In this video, the JS-BOX is mounted on an industrial cleaning machine.
It evolves in a 20,000m2 supermarket with its camera fully turned towards the ceiling.
It uses a single RGB camera 640×480.
Video is accelerated for convenience.
After having done 400m, JS comes back to a zone already mapped:
We can notice at this moment that the precision before this loop closure is excellent (about 10m difference) and that the map becomes consistent once the loop is closed.
WHAT MAKES JACK SLAM DIFFERENT FROM OTHER POSITIONING PRODUCTS
We have worked hard to make our system robust and reliable. In this sense, it is unique in multiple ways:
It performs data-fusion with other sensors (even with a LIDAR).
It can relocate in less than a second (for example if we hide its camera, even for a long time).
It continues to work perfectly even if the environment has completely changed (for example a warehouse has been totally restructured).
It needs little to provide efficient localization: A simple 4-core CPU and a low-cost camera.
WHERE TO USE IT
Wherever an accurate indoor location is needed.
To locate any moving element in 3D within a few centimeters.
Logistics and Assembly plants.
Public places (hospitals, museums, airports, …).
Supermarkets and Retail stores.
Exhibitions and Shows.
Hotels and Restaurants.
SLAM (Simultaneous Localization And Mapping) is the way to locate a device indoor.
The ability to build a map and simultaneously localize within that map is essential in all robotic applications. The input sensor for SLAM is traditionally a LIDAR.
Visual SLAM (VSLAM) uses a mono or stereo camera as input to localize, so a low cost sensor.
JACK SLAM is a VSLAM.
This technology, much more efficient than LIDAR in many cases, is old but has never been able to have a practical use because of the important CPU resources required. Today, both powerful and inexpensive hardware and the optimizations made by ip sum tek allow the use of VSLAM in real time on low cost computers.
The principle to locate is like the one used by humans: Detecting the natural features seen by the camera (a difference in color, an angle, a shape, …). Have a look to the videos of this site to understand the principle.
The easiest way to test the JACK SLAM positioning system is to purchase a JS-BOX-DEMO.
This box, with calibrated camera, computer and battery is ready to use. You can use it handheld and walk around, it will locate you. Or you can temporarily install it on any mobile system (a better way to test it in-situ).
The location is transmitted via Wi-Fi and you can see the result immediately on a PC with the supplied software (the result is the map built in real time).
There are 3 systems that can be integrated:
– The JS-BOX, which is a small case containing a computer, the software and one full license. With an extended power supply range, it is provided ready to be mounted on the robot. This box will provide the robot’s position by Wi-Fi or Ethernet.
– The JS-BOARD, a small and low-cost board uncased wich is a small stand-alone computer. The software and one full license are included. It is provided ready to be integrated in an existing archistecture. This board must be powered with 5VDC and communicates via its Ethernet port.
– JACK SLAM is a software that can be integrated on any Linux-based computer as a third-party sofware. We propose an integration pack including licenses and support. This pack permits to quickly get an operational localization.