INNOROBOHACK (Innopolis, November 30 - December 1, 2019)
Experts from the Center for Technology of Components of Robotics and Mechatronics of a Russian IT university together with JSC NPO Android Technika and JSC NIIAS (JSC Russian Railways) prepared tasks for two tracks - anthropomorphic robotics and autonomous transport.
1. Anthropomorphic robotics.
Robot capturing and moving an object in a simulator environment.
The task is to perform the Fedor robotic complex actions in the simulator environment at the ISS station. The image obtained from the cameras of the physical sample of the robotic complex must be obtained using the mjpeg protocol, processed and the control effect on the robot transmitted using the control protocol.
The operator can control the engines responsible for the movement of manipulator modules, head and case. Motor control is possible by position or by moment. The pedipulatory modules are fixed and cannot be controlled. To complete the task, it is necessary to make manipulator modules of the robotic complex move a given object located in the field of view of the cameras of the head module to the specified area of space.
In the course of the task, it is required to carry out the capture and movement of an object in several locations. Participants can have completely different skills, preferably knowledge of the basic deep learning frameworks, experience with computer vision tasks and reinforced learning.
2. Autonomous transport.
Determination of rail track by deep learning methods.
The goal is to develop a technology for automated control of the train without the participation of the driver. One of the key tasks for achieving this goal is the task of searching for a rail track using data provided by machine vision cameras.
The task of semantic segmentation is quite popular and did not pass our project. It is the search for the route of the train in the future that allows you to identify areas of interest for the detection of infrastructure.
Participants are provided with a dataset consisting of more than 4,000 manually marked railroad tracks of the Moscow Central Ring, obtained using a front camera mounted on the front of the locomotive. Binary image masks are used as annotations.
To evaluate the quality of the algorithms, we recommend using a metric recognized in the international community when solving problems of semantic segmentation - IoU. Participants can have completely different skills, preferably knowledge of the main deep learning frameworks, experience with computer vision tasks.
The prize fund of the hackathon is 360,000 rubles.
Applications are accepted until November 29 from teams of 2-5 people. Within 48 hours, participants will have to develop a product prototype in one of the selected areas. Mentors and experts from customer companies are involved in the development. After this, the teams will present their decisions to an expert jury.
The winners of each direction will receive 100,000 rubles for 1st place, 50,000 rubles for 2nd place, and 30,000 rubles for 3rd place.