You Fan: On the eve of the dawn of unmanned vehicles, Baidu's Apollo platform is upgraded again

January 8, local time in Las Vegas, USA, is the day before the official opening of CES 2018. At 2:53 p.m., Baidu just finished announcing version 2.0 of its Apollo program. At this moment, at 6:53 a.m. Beijing time on January 9, 10 cars equipped with Apollo 2.0 version drove slowly within the camera at Baidu's headquarters campus. These vehicles include both Lincoln models and Golden Dragon buses, as well as unmanned delivery vans. They advance noiselessly in the cold wind, waiting for the moment when the dawn comes.

This is the site of the China-US link at the Baidu World Conference in Las Vegas. During the 3:10 minute video link, the vehicles demonstrate parading forward, turning around and other maneuvers. One of the buses without a steering wheel had several Baidu employees riding in the passenger seat. Another small car had no driver in the driver's seat and the vehicle swung the steering wheel on its own, completing a smooth turnaround.

"This link-up includes both a demonstration of the results of our partners and products based on the same platform and geared towards different experimental scenarios." After the meeting, Li Zhenyu, general manager of Baidu's autonomous driving division, said with slight excitement, "Signal stability was a huge challenge during this Sino-US connection - it was windy in Beijing and cold after 6 a.m. "

This fleet of vehicles is the result of the Baidu Apollo version 2.0 capability module. In the six months since it was first announced, the Apollo platform has finally achieved full openness of the capability modules and further enhanced some of its core capabilities after two version iterations. This corresponds to the continued growth of the list of partners.

So, will such an Apollo platform lead eco-partners to the dawn of unmanned cars?

Capacity upgrades

The upgraded Apollo 2.0 version illuminates four modules including cloud services, software platform, reference hardware platform and reference vehicle platform, which are capable of simple urban road autonomous driving. This also means that the Apollo platform's capability modules are all open to the public.

"Compared to previous versions, Apollo 2.0 opens up the ability to recognize traffic lights, change lanes, overtake, and for obstacles on the road, classify and identify them."

Wang Jing Ao, head of research and development for Baidu's Apollo platform, said on Jan. 6, local time in the United States.

According to Wang Jing Ao, this Apollo 2.0 release opens up safety services for the first time and further strengthens capabilities such as self-location, perception, planning and decision making and cloud simulation. One of the self-positioning capabilities can accurately detect the environment, ensuring that the car can operate in urban, canyon, tunnel and other environments. The planning decision capability enables vehicles to accurately identify traffic lights and intersections, turn faster and more accurately and avoid obstacles.

The sensing module, on the other hand, helps vehicles improve traffic light screening to 99% and enables obstacle recognition within a range of 500 meters. The simulation capability has enabled data uploads from 30 minutes in the past to 30 seconds today, and includes realistic road simulation and one-click recall capabilities.

"Many companies, including Baidu, are focusing on autonomous driving based on mobility services, and this will involve a wide variety of driving scenarios." Roy Wang, senior automotive analyst at IHS Markit, commented to 21st Century Business Herald, "In Apollo version 2.0, Baidu has strengthened the capabilities of its perception and decision-making algorithms. These help expand the range of scenarios in which their solutions can be applied, thus helping the transition to complex scenarios. "

According to Roy Wang's analysis, in the future consumers will embrace self-driving mobility services with relatively mature technology, and regulations will gradually evolve with the deployment of autonomous driving. "With these two points and a few other elements in mind, we predict a lower but steady rise in production of self-driving cars over the next decade. After 2040, autopilot will see a faster growth and higher production. "

just before dawn

Although the capabilities of the Apollo platform continue to strengthen, and agencies generally predict a steady future for the unmanned vehicle industry, the industry's technical bottlenecks remain insurmountable from now on.

Baidu's unmanned vehicles are quite perceptive of obstacles, generating braking behavior almost the moment the traffic light turns from green to yellow, and even sometimes stuttering as a result. Multiple Baidu sources told the 21st Century Business Herald that the perception and decision-making capabilities of Baidu's unmanned vehicles are in the terminal vehicles.

"In the current Apollo platform, auxiliary capabilities including high precision maps, simulation tests, OTA wireless upgrades and DuerOS system fall in the cloud, and the rest including sensors and software layers fall on the terminal vehicle."

Li Zhenyu said.

Among them, the software layer includes several capability units such as perception, localization, path planning, control, and interaction, as well as the complete system software framework and underlying real-time autonomous driving system formed on this capability unit. In other words, the "brain" that perceives and reacts to various road conditions during the driving of an unmanned vehicle is inside the vehicle.

This also means that within a moving unmanned vehicle, the system needs to perform a lot of data calculations based on real-time road conditions. In a driverless car, the camera generates 20-60MB of data per second, the radar generates up to 10kB per second, the sonar 10-100kB per second, the GPS will run at 50kB per second, and the LIDAR will run between 10-70MB per second - each driverless vehicle will generate around 4TB of data per day.

The large amount of data calculations performed at the terminal will make the energy consumption ratio inside the vehicle inevitably high. "Reducing the energy consumption ratio is what the industry needs to work together on." A current trend in the industry is to load trained models onto terminal chips for application, but the best solution is still the expansion of terminal processing computation," said Xiang Yang, deputy general manager of the Artificial Intelligence Industry Research Center at Sedi Consulting. "

Li Zhenyu, on the other hand, admits that the current energy consumption problem on the terminal is a problem that is difficult to get around in the productization of unmanned vehicles.

"But what needs to be addressed more than anything else at this point is safety," he stressed, "for example, the recognition of traffic lights is extremely difficult for machines, with situations such as backlighting, multiple lights, and same-colored obstacles being challenges that machines will face in accurately identifying traffic lights, which requires a lot of training and a lot of investment in product safety. "

As for the current specific energy consumption ratio of Baidu's unmanned vehicles, Li Zhenyu said he could not reveal it for the time being, "different vehicles have different power consumption." However, he remains confident that "the Apollo platform architecture is relatively reasonable in terms of the current situation, and future terminal capabilities may change with the introduction of 5G technology, where models based on 5G technology may be placed in the cloud," he said, adding that "the issue of low power consumption as well as low cost will be addressed when it comes to mass production of unmanned vehicles. "

1、Service Provider SDNNFV Revenues to Reach 22 Billion by 2020 Report Says
2、Swoolebased WeChat Sweep Login
3、Elderly living alone in need of care and attention Robots show their hand
4、China Aviation Industry International Big Data Nuggets
5、Showcasing the Future of Mobility RinspeedSnap Concept Unveiled

    已推荐到看一看 和朋友分享想法
    最多200字,当前共 发送