ETU “LETI”: Project for Maintaining Self-Driving Cars Online
Today, robotic systems are in great demand in all areas. Not only are they capable of detecting and preventing fire hazards, creating millions of parts for machines and equipment, but also of transporting humans from one location to another. Launching self-driving solutions not in the simulation but in the real world poses great problems of controlling these vehicles and quality assessment. Therefore, it is crucial to maintain these systems in a timely and responsive manner, but operators do not always have access to the equipment. “Access to equipment could always be an obstacle to the development of science, and that was particularly well demonstrated in the pandemic,” notes Konstantin Chaika.
The project by Konstantin Chaika, a graduate student of the Department of Software Engineering and Computer Applications at LETI, focuses on remote debugging robotic equipment and expanding the possibilities of working with driverless vehicles.
“The idea of the project is to develop a service for a remote run and debug various control algorithms for self-driving robots. In other words, through this service, anyone with only a computer and the Internet will be able to remotely access a real robot, run their program on it, and get a detailed result, including online broadcasting.”
Konstantin Chaika, a graduate student of the Department of Software Engineering and Computer Applications at LETI
The platform involves a special lab simulating a smart city environment. Despite its reduced size, the simulation is sufficiently accurate for research projects, and its monitoring systems allow quantifying the quality of solutions. Such a lab is particularly relevant for the Sim-to-Real approach, one of the key trends in self-driving vehicle control based on AI. This approach consists in developing a solution in a simulator that can be transferred to the real world with the same behavior and performance.
“The invention can be seen not only as a ready-made solution but also as a parameterized model, which provides the possibility of remote debugging of developed algorithms. It can be adapted to different environments or types of robots, allowing to solve not only the problems of self-driving transport in a smart city environment but also navigation in an industrial environment. That will make it possible to solve tasks from logistics to controlling manipulators or other industrial robots.”
Konstantin Chaika, a graduate student of the Department of Software Engineering and Computer Applications at LETI
The graduate student is confident that this project will greatly expand the possibilities for developers of programs for autonomous robots. All they will need to run their developments on a real robot is access to the Internet. Not being able to build a lab or go to an existing one will no longer be a barrier to science. “In addition, this is a great platform for remote competitions, allowing anyone to run their solution, get detailed results and compare it with other participants solving similar problems,” commented Konstantin.
The main analog of the project is the system for validating the solutions of the AI Driving Olympics, based on the Duckietown project. However, launching solutions using this system has limitations and does not allow accessing the robot for an extended time, but only to launch a program within one of the Olympics stages.
The project has two stages. The first step is for the developer to create a platform mockup at the Intelligent Transportation Systems Laboratory at the Alexander Popov International Innovation Institute for Artificial Intelligence, Cybersecurity and Communications. It will contain the location of roads, monitoring systems and other peripheral equipment, as well as design the physical structure of the city. The final stage will be devoted to the development and debugging of a prototype of monitoring and resource management software.