The companies have jointly developed an AI robot control system that can interact with the physical world and be used in various fields from logistics to rescue operations. Tests have shown that in real conditions the system can operate with an accuracy of up to 95%. The research project has no analogues in Russia.
Moscow, 7 December 2020. — Sber and Microsoft Research summed up the results of a joint project launched in October 2019 to develop a unique AI system that offers the opportunity to teach robots to manipulate physical objects of unstable shape in almost the same way that humans do. A key goal of the research project was to develop a robotic system that could collaborate with human operators in changing environments, freeing teams from performing routine work and ensuring greater safety throughout the operation. In addition, this is one of the first examples of transferring such technology from simulation to the real world — the trained system has been implemented and tested in Sber’s Robotics Laboratory.
A video with a demonstration of the robot’s functionality is available here.
“Together with Microsoft, we managed to achieve results that are very important for the development of artificial intelligence and robotics. The system can be used in many sectors of the national economy. We intend to continue research in order to promote the development of digital technologies and provide Russian companies with modern high-performance robots,” said Stanislav Kuznetsov, Deputy Chairman of the Executive Board of SberBank.
Work on the project was carried out at Sber’s Robotics Laboratory in Moscow, Russia, in collaboration with Microsoft researchers in Berkley, California, and Redmond, Washington, in the U.S., and collectively lasted over a year from May 2019 to May 2020. The research team consisted of 8 people. Their practical objective was to develop technology using a robot-manipulator to unload collection bags with coins weighing up to 6 kilograms from mobile carts for further processing in counting machines. In contrast to gripping solid objects, manipulating unsteady objects with a moving center of gravity requires constant calculation of the gripper’s position and orientation in each particular case. Methods of deep learning and reinforcement learning were used to calculate and predict these parameters. Tests showed that bags could be unloaded successfully under real conditions on the first attempt in over 95% of cases.
Work on the project was carried out in three stages. The first stage was the development of a realistic simulation of the robotic unit and the objects of manipulation — deformable bags with coins and their physical characteristics. The second stage was integration of simulators with the machine learning framework and its reinforcements and conducting experiments in the simulation environment to train an intelligent agent to control a virtual robot. For this purpose, the researchers applied the machine teaching approach. That technique involves algorithms not only learning by interacting with the environment, as in classical reinforcement learning, but also by using “human prompts”. At this stage, it was necessary to prescribe correct reference points for the algorithm that it could learn from. This makes the process much faster and more efficient. In the final stage, the intelligent agent trained in the simulators was transferred directly to the physical robot. As a result, Sber’s Robotics Laboratory successfully tested real manipulations of the robot unloading sacks of coins from a cart.
The system creators are confident that such technology can be applied to a wide range of areas. It is especially useful for tasks where a person is exposed to danger or have to make extraordinary physical efforts, for example, in rescue operations, oil production, logistics, etc. — everywhere where it is necessary to perform actions with objects of unstable form.
“We highly appreciate the results of the joint research project and are happy to continue our cooperation with Sber in this and other areas,” said Kristina Tikhonova, president of Microsoft Russia. “The level of trust that has been established between our companies allows us to make a significant contribution to the development of applied AI in Russia and worldwide, as well as to contribute to the digital transformation of business and economy in our country.”
The companies intend to continue their cooperative research to promote the development and wider distribution of products at the intersection of robotics and artificial intelligence.
About PJSC Sberbank
PJSC Sberbank is Russia’s largest bank and a leading global financial institution. Holding almost one-third of aggregate Russian banking sector assets, Sberbank is the key lender to the national economy and one of the biggest deposit takers in Russia. The Government of the Russian Federation represented by the Ministry of Finance of the Russian Federation is the principal shareholder of PJSC Sberbank owning 50% plus one voting share of the bank’s authorized capital, with the remaining 50% minus one voting share held by domestic and international investors. Sberbank has customers in 18 countries. The bank has a major distribution network in Russia with about 14,000 branches, while its international operations – subsidiary banks, branches, and chapters – include the UK, US, CIS, Central and Eastern Europe, India, China, and other countries. It holds general banking license No. 1481 dd. August 11, 2015, from the Bank of Russia. Official websites of the bank: www.sberbank.com (Sberbank Group website), www.sberbank.ru.
On September 24, 2020, Sberbank underwent a rebranding, offering financial and non-financial services of the bank and Sberbank Group to individual and corporate customers. Today, the Sber ecosystem is a raft of services for life and daily assistance in handling pressing everyday issues for individual customers and businesses. The Sber ecosystem website: www.sber.ru.
Contact information:
Press service
About Microsoft
Microsoft (Nasdaq “MSFT” @microsoft) enables digital transformation for the era of an intelligent cloud and an intelligent edge. Its mission is to empower every person and every organization on the planet to achieve more.
Contact Information:
Microsoft, 125252, Moscow, PO Box 70
Tel.: 7 (495) 967-85-85
Fax: 7 (495) 967-85-00
Latest Microsoft news can be found at the Microsoft Russia news center.
Media representatives can also obtain additional information from PRP Group – a Weber Shandwick Affiliate Company at [email protected] or by calling (495) 937 3170.