LETI: LETI Researchers Know How to Improve the Performance of Autonomous Robots
LETI researchers suggested an algorithm for filtering laser scans from lidars, matching them to each other. The new approach will reduce the average processing time and improve the robot performance by optimizing the algorithm. The scientists developed formulas that enable filter adaptation to a particular robot with known speed and characteristics of lidar and suggested using an indoor corridor detector. The study results are published in the Robotics and Autonomous Systems journal.
Algorithms that use sensors, such as SLAM (simultaneous localization and mapping), SfM (structure from motion), and others, are highly demanded in modern robotics. Such sensors have a common disadvantage – they simultaneously collect too little and too much data. On the one hand, there is too little data because it is impossible to smooth or approximate it without a significant loss of accuracy. At the same time, there is too much data because it requires a lot of memory to store and process scans that appear every 30 ms. There is no need to capture laser scans so frequently unless such laser scanner is mounted on a car moving at 60 km/h since such amount of dense point clouds from laser rangefinder is excess.
Having analyzed the existing methods of reducing the size of laser scans, Kirill Krinkin, Head of the Department of Computer Science and Engineering at LETI, and Anton Filatov, Assistant Professor of the same department, concluded: to speed up filtering, that it is necessary to reduce the amount of input data and determine the value of each scan for the subsequent transfer of an original scan to a core of a SLAM algorithm.
To solve this problem, LETI researchers suggested using a correlation filter for 2D laser scans. Its core idea is to compare the current incoming laser scan to the previous one. If the latter is ‘similar,’ then the current scan should not be processed. “To avoid the noise in observations, it is better to compare an incoming scan to several previous scans. Hence, a sliding window of scans appears that plays a role of reference for new incoming scans,” explains Kirill Krinkin.
To calculate the scan correlation, it was proposed to create histograms for each of them based on a division by ranges and angles and then compare them by calculating the Pearson correlation coefficient.
In order not to lose useful information when the robot moves through featureless environments, such as corridors, scientists proposed to detect corridor scans before filtering and not drop them. This requires looking through every point of a laser scan and calculating the sign of range difference of the current point and the next one.
The researchers conducted experiments on MIT and TUM datasets for vinySLAM, Gmapping, and Cartographer.
“Experiments show that the suggested filtering algorithm works incomparably faster than a scan matcher algorithm (5.9⋅10−5 s for filtering, 12.9⋅10−3 for scan matching of vinySLAM). Therefore, the filtering process, in general, saves computational resources because the time saved from scan matching might be used for other possible needs of a robot. The application of this filter reduces the average scan processing time by more than 40%.”
Kirill Krinkin, Head of the Department of Computer Science and Engineering at LETI