Robotics & Automation 2019
Ken Smillie will be speaking at the Robotics and Automation 2019 show at 10.30 on 20th October.
USA Payment Plan
UK payment plan info
Some SLAM software algorithms have been made available as open-source on the internet, but they are purely algorithms and not a product that you can take and use off-the-shelf. SLAM is most successful when it is tightly coupled and designed with specific hardware in mind. A generic SLAM cannot perform as well as one that has been specifically designed for a purpose.
Visual SLAM is closer to the way humans navigate the world, which is why it’s popular with robotic navigation. But in the same vein, vSLAM will have the same image-capture challenges as humans do, for example not being able to look into direct sunlight, or not having enough contrast between the objects picked up in the image. These can be overcome indoors, however, you may need to map a forest, tunnel or urban canyon. While SLAM technologies don’t rely on remote data (meaning you can scan areas where there is no GPS), you do need to ensure the SLAM technology you chose operate well inside, outside, in daylight and darkness.
Mapping a property is time-critical. Ideally, you want to make a single visit and gather sufficient data to create a highly accurate 3D model. Ensure the software you choose transforms 3D point cloud data into actionable information in real-time. This allows you to view and interrogate your data whilst still in the field, and make any adjustments, or collect missed data, then and there.
If you’re trying to map an enclosed environment (e.g. tunnel, mine) or a complex, difficult-to-access space such as a heritage building with tight stairwells and uneven floors, you need to use fully-mobile, adaptable technology. Wheel-based systems, often used with the vSLAM camera, will struggle with access. Handheld devices or LiDAR scanners that can be attached to a drone or pole and still deliver accurate results in a rugged environment are best for navigating hazardous spaces.
While vSLAM is able to provide a qualitative high-level map and sense of the surrounding features, if you’re needing survey-quality accuracy and rich-feature tracking at a local level, you’ll need to consider LiDAR. Cameras require a high-frame-rate and high processing to reconcile data sources and a potential error in visual SLAM is reprojection error, which is the difference between the perceived location of each setpoint
and the actual setpoint.
In order to deliver the depth required for high-quality data, a number of depth-sensing cameras are needed with a strong field of view. In most cases, this isn’t possible, especially as cameras with high processing capabilities typically require larger batteries which weigh down airborne scanners, or limit the time of flight. LiDAR is both faster and more accurate than vSLAM, and can deliver detailed point clouds without expensive (and timely) camera processing.
