Archives 2017

Man Vs. Machine: Russian Company Compares Human, AI Driving Capabilities

Man Vs. Machine: Russian Company Compares Human, AI Driving Capabilities

Link to publication

A Russian software developer called Cognitive Technologies has conducted a detailed study comparing AI reaction to that of human drivers. Speaking to Sputnik’s sister agency, the company said its findings demonstrate that AI capabilities to operate a motor vehicle are rapidly expanding, coming close to matching or even surpassing those of people.

The Moscow-based company, creator of its own automated driving assistance technology known as Cognitive Pilot, has completed a study to determine who – human beings or artificial intelligence, are better able to detect road signs, other vehicles and pedestrians in various traffic situations.

Speaking to Russia’s RIA Novosti, Cognitive Technologies’ autonomous software department head Yuri Minkin said that the tests’ main goal was to determine whether its AI had already outmatched human drivers’ reaction time and accuracy.

Put up against 17 human volunteers, the company’s AI was put to the test in a variety of challenging conditions (dusk, nighttime driving, driving in rain or conditions of blinding sunlight). The volunteers and the AI took turns identifying a series of objects on the road on a single monitor.

Taking place between September and November, tests featured the use of a single camera, and were conducted along 27 different urban routes at speeds between 50-60 km/h. Researchers limited the number of objects appearing onscreen simultaneously to three for the human volunteers’ benefit.

The first stage, studying the quality of detection of objects in good weather and road conditions, showed humans and AI as near-equals, in terms of both accuracy and speed, exceeding 99% accurate detection rate.

Meanwhile, the AI proved superior when it came to objects that were not completely visible, hidden behind trees, parked cars or other obstructions. The AI also proved faster in detecting road signs, picking up even obscured signs in just a fraction of a second.

“The testing showed that under more difficult conditions, the human volunteers often noticed the road signs a moment later than the AI. This time gives the [AI] control system an additional advantage for the processing and analysis of information about the situation on the road as a whole,” Minkin said.

The AI was slightly better in terms of speed and accuracy of detecting objects in rainy conditions, (98.3% vs. 97% for humans). In twilight and blinding sunlight, human volunteers proved significantly slower in recognizing road objects, although the overall recognition rate remained 98% for both.

One area where AI lagged behind its human counterparts was in spotting pedestrians in rainy and nighttime conditions (with an accuracy of 98.2% compared to 99.2% for people shown).

“A person is a rather complex object for recognition,” Minkin explained. “Pedestrians do not have a constant form, and can travel embracing, holding hands, carrying an object, etc. And while the AI can match the abilities of a [human driver] in clear conditions, in more difficult conditions, the human’s abilities are still slightly superior.”

According to Minkin, the main takeaway from the study is that artificial intelligence is rapidly approaching the capabilities of human beings in the recognition of objects on the road. Furthermore, the more complex the road conditions, the better the AI performs, relative to its human counterparts.

“It can be expected that with an increase in computing power, and in the quality of sensors and software, the advantage of the AI will become all the more obvious, similarly to how it has become in chess,” he concluded.

Cognitive Technologies president Olga Uskova says that the company has plans to conduct further testing, and plans to use industry experts to come up with a more advanced testing methodology.

“But we must understand that this was the first attempt to compare the capabilities of artificial intelligence and people. With this first approximation, we’ve received real results, which can and should be considered, and on whose basis it will be possible to predict the direction in the creation of AI that can drive vehicles,” she noted.

Russian autonomous harvester works into the night

Russian autonomous harvester works into the night

Link to publication

Russian IT-company and Ai developer, Cognitive Technologies, has released a new version of its autonomous driving system for harvesters that can work at night.

The latest development follows the company’s successful field testing of a prototype autonomous harvester in August (see Grain Central story).

Cognitive Technologies president, Olga Uskova, said the ability to work at night time would minimise the impact of the weather and human factors and could improve the quality of grain harvesting by 25-30 per cent.

She said the new system was capable of performing more complex manoeuvres than its predecessor. “At any time of the day the harvester equipped with the system is able to make a U-turn or move along a difficult and curved trajectory,” he said.

“The maximum speed of the autonomous harvester now is 11 kilometres/hour, while it automatically monitors the edge of the collected crop.”

The system itself determines if the harvesting collection boundary starts to deviate from a straight line. If this happens, the combine itself simply corrects the track and continues harvesting.

“It is important that the harvester is able to determine not only the edge of the field: the system recognises other machinery, people and animals,” Mrs Uskova said. “It also understands where the field is already processed, and where it is not.

“In future the number of objects for detection and segmentation will be increased and this will make it possible to distinguish different cultures growing on the same field.” Mrs Uskova said the machine used a camera as its main sensor for collecting data and navigating.

“The harvester does not need a high-precision GPS: it can be completely autonomous, relying only on the video signal,” he said.

Tests were carried out on the TORUM 760 combine harvester jointly with the Rostselmash combine manufacturer in the Rostov region of southern Russia.

The autonomous harvester worked on a rice field of 40 hectares and successfully passed all the tests in the night. Cognitive Technologies expects to create a fully autonomous grain harvester by 2023-2024.

Fully driverless combine harvester by 2024

Fully driverless combine harvester by 2024

Published on AGRILAND.IE
Link to publication

An off-the-shelf combine harvester has been converted to automatically steer itself by a Russian technology company, at a “fraction of the normal cost of such technology”.

This, says the company, is a precursor to fully autonomous (driverless) operation in the near future.

The company is called Cognitive Technologies; it develops robotic systems and software.

What are the advantages of such developments? The company claims that the use of such technology can increase output and reduce losses. It allows the combine harvester operator to “focus on machine settings, rather than steering and driving”.

Working in “low-growth” barley (standing just 30cm tall), the challenge was to come up with a practical system using only one video camera, instead of a plethora of cameras and sensors. In effect, the company says that its system employs only one camera and a “neural network”, unlike “much more expensive” systems from mainstream machinery manufacturers.

According to the company, the system works to a high level of accuracy – targeting a sub-5cm tolerance. The current system automatically stops the combine harvester upon completing each task – awaiting intervention by the driver. A task could encompass multiple rounds of a field.

Olga Uskova, President of Cognitive Technologies, explained: “We’ve spent five years developing this technology thus far. We expect to create a fully autonomous grain harvester by 2023-2024.

“We have already created a highly-developed computer vision system that allows just one camera to achieve similar results as leading international manufacturers get with three to four sensors. As a result, the cost of the equipment is three to four times lower. And this gives us a significant competitive advantage,” she added.

The company field-tested its equipment during recent weeks; the system was fitted to a Russian-built Rostselmash RSM 181 Torum harvester. The tests were carried out in the Rostov region of southern Russia.

Over the course of one and a half days, the system was “re-trained” from wheat to barley. At present, it recognises the edge of the crop, thus determining where the machine should steer itself. According to the company, it can do this “better than a human driver”.

The camera and associated software can identify all manner of obstacles including: uncut versus cut areas; swaths or layers of straw; trees; roadways; people; etc.

Russians test driverless harvester

Russians test driverless harvester

Link to publication

Russia has stepped into the era of automated machinery with a Russian company field testing an autonomous grain harvester.

Cognitive Technologies, which develops robotic systems and software for machine vision, has field tested a Rostselmash RSM 181 Torum harvester in the Rostov region of Southern Russia.

Company president, Olga Uskova, said the company had spent more than five years developing the system and expected to have a fully autonomous grain harvester available for commercial release by 2023-2024.

She said the system was low-cost because it used a single video camera as a sensor, unlike other systems that used laser scanners, stereo cameras, lidars (light detection and ranging), etc.

“The automated complex of our smart harvester includes a video camera and artificial intelligence system,” she said.

“The neural network of the system is now able to allocate five main classes of the field scene:

1) Unmowed part of the field

2) Mowed part of the field

3) Roll in the hay

4) Everything that falls in the view of the camera from the header of the harvester

5) All other objects – obstacles, other machinery, forests, trees, roads, weeds, people etc.

“The additional competitive advantage of our Ai system is that it sees various types of field boundaries (edges, rolls).”

Mrs Uskova said in the field tests, instead of the planned field of wheat that would have been more than 80 centimetres high, farmers provided fields of barley that were about 30 cm high.

“This was a great, almost impossible challenge. For 1.5 days we retrained the neural network based on the datasets of other agro cultures. As a result, the neural network was able to determine the stacked roll and the field scene better than the person. The result surpassed all expectations. This is a breakthrough,” she said.

VisionHack – First international hackathon on computer vision for unmanned vehicles

VisionHack – First international hackathon on computer vision for unmanned vehicles

Published on SOCIABLE.CO
Link to publication

Recently, the words “Russia” and “Hack” have been paired together in the news with some pretty negative stigma, however VisionHack presents a different narrative regarding hacking in the Russian capital.

VisionHack will be the first university hackathon focusing on artificial intelligence and computer vision with prizes of more than $30,000.

The Moscow based competition will take place from September 11th-13th and will attract talent from around the world such as teams from MIT (USA), the University of Cambridge (UK), the University of Science and Technology Beijing (P.R. China), UPC (Spain) along with a number of university teams from Russia.

The competition will be jointly organized by NUST MISIS, one of Russia’s leading technology universities, and Cognitive Technologies, a company which develops AI systems for unmanned vehicles.

According to Alevtina Chernikova, Rector of NUST MISIS: “It is not a coincidence that Russia and NUST MISIS have become a venue for holding such high-level events-Russia has one of the world`s best schools for the development of artificial intelligence. Today, our domestic developments in the field of computer vision are highly rated by leading international experts, and the NUST MISIS team for several years in a row has reached the final of the ACM/ICPC international programming championship”.

The participants’ task will be to create their own intellectual subsystem for driver assistance and ADAS (advanced driver assistance system) capable of automatically detecting various events of the road such as: a car stopped in the middle of the street with its hazard lights on; or a dog or small child about to cross the street, essentially anything that might pose as a conventional danger within a street setting.

“Participants will have to solve real and significant practical issues of computer vision for unmanned vehicles”, said Olga Uskova, president of the Cognitive Technologies group.

The team which presents the most accurate and comprehensive solution to the set of proposed road situations will be crowned the winner and secondary prizes will be awarded for qualitative decisions on specific developments.

This is the first hackathon in the world dedicated to computer vision technologies and unmanned vehicles. Within recent years, the focus on technology related to unmanned vehicles has been a point of interest for tech giants and governments everywhere such as Lyft and Uber, who have announced ambitions to advance their own self-driving technology. Moreover, experts expect self-driving technology to be one of the biggest technological advances by 2021, making the need for this competition all the more relevant.

Russia's Self-Driving Car Company Is Coming for the World

Russia’s Self-Driving Car Company Is Coming for the World

Published on INVERSE.COM
Link to publication

Can a team of Moscow developers take on Tesla and Mobileye?

A mysterious self-driving car company has been quietly expanding in recent years in the world’s largest country. Now, Moscow-based Cognitive Technologies has hired a slew of new recruits and is ready to move to the U.S. in the coming months.

“The big R&D center will stay in Russia, but the main engineers and business guys will be sent to U.S. soil to set up a proper office,” Roman Tarasov, the company’s VP for global business, tells Inverse. “We’re thinking either California or Delaware.”

Cognitive Technologies was founded in 1993 by the guys who created Kaissa, the world’s first computer chess champion. For decades it worked on image and voice recognition applications, selling products to Intel, Yandex, and others. In 2014 it announced Cognitive Pilot, a program to develop software for self-driving cars, and in 2015 it announced a partnership with Russian trucking giant Kamaz — with plans to roll out assistive driving technology in 2017 and self-driving trucks by 2020. Last year, the company revealed a self-driving tractor.

In the past year, the company has expanded rapidly. The whole company, entirely based in Russia for now, consists of 1,500 people. In the past few months, Tarasov says the C-Pilot team alone has almost doubled, reaching about 90 programmers plus 25 people on the business team.

“We’re in the middle of a process to become well-known in the English-speaking media,” Tarasov says.

Little is known about C-Pilot’s technology besides what the company has said itself. The company, not surprisingly, claims it has a breakthrough, capable of matching the Tesla’s Autopilot program, Mobileye’s self-driving software, and every other company working on an autonomous car.

Unlike competitors such as Mobileye, which uses proprietary chips, the developers tout C-Pilot’s ability to run on standard computer hardware, making it flexible and easily upgradeable. C-Pilot can supposedly do this thanks to a superior ability to focus on important objects in a scene, allowing it to make do with less computing power.

It’s also supposedly better at dealing with bad roads and bad weather.

“The whole project was built to design the autopilot for real roads, Russian roads,” Tarasov says. “Most of the roads on the planet are like this. So lack of light, snow, fog, bad road marks. If it works in Russia, it will definitely work in Brazil. So that’s the idea, to build an autopilot for real roads.”

The system can currently steer, push the accelerator, emergency stop, and warn about traffic lights. By the end of the year 2019, the team wants to achieve level four autonomy. Although still under development, the system can supposedly recognize 36 road signs with over 98 percent accuracy.

A big part of moving to the U.S. is the potential to get more funding. Thus, Tarasov says, the company is looking to locate near key investment offices and to make a splash.

Currently, Cognitive Technologies develops its system using a Nissan X-Trail and works with car makers to test the software on their own vehicles. Whatever they use, the company has its eye on launching a fleet of five-to-ten autonomous cars, whizzing around near investors.

“If Google has one, why shouldn’t we!” Tarasov says.

When Inverse asked Tarasov to clarify what makes C-Pilot different from Tesla’s Autopilot, he said, “Tesla kills people and we don’t!” He was, apparently, referencing the May 2016 death of Joshua Brown, who was killed while driving a Tesla Model S in Autopilot mode — though we’ll note that the National Highway Traffic Safety Administration did not find a safety default in Tesla’s systems during an investigation of the crash.

If Cognitive Technologies breaks into America, it will be a big step forward, but it will also face more competition than ever. In California alone, there are already 27 companies testing their own self-driving cars.