A huge set of neural networks was developed and trained for technical vision systems, which were subsequently optimized to build a dynamic semantic model of the scene in real time.
MOSCOW, October 22. /tass/. The newest Russian robot "Marker" has received all the most modern modules that give it real machine vision. Boris Vishnyakov, Deputy head of the Department for Software Complexes of GosNIIAS, associate Professor of the Department 806 "Computational Mathematics and Programming" of MAI, told about this during a speech at the School of Management of the Moscow Aviation Institute.
"We have developed modules for working with equipment, a module for recognizing objects in 2D and 3D, a module for semantic segmentation, a module for calculating depth, a module for automatically localizing ourselves in space and plotting the trajectory of our movement, a module for calculating a semantic cross-country map, as well as modules for autonomous movement and motion control for several platforms have been developed by our colleagues in cooperation," he said.
According to him, in the process of creating a "Marker" for vision systems, a huge set of neural networks was developed and trained, which were subsequently optimized to build a dynamic semantic model of the scene in real time, so that the platform understands which objects are around it in order to optimally drive along a given route.
The experimental robotic platform "Marker" was developed as part of a joint project of the Foundation for Advanced Research and the NGO "Android Technology", launched in 2018. The goal of the project is to create and conduct a full-scale development of technologies and basic elements of ground robotics.