Auto Tech

Open source Apollo speeds up Baidu's self-driving software development

The Chinese technology company announced version 1.0 of its open-source Apollo software in July. Three months later it's up to version 1.5.

Baidu President Ya-Qin Zhang at Apollo meetup

Baidu President Ya-Qin Zhang announces Apollo 1.5 at a meetup in Sunnyvale, California.

Wayne Cunningham/Roadshow

In July, Chinese technology company Baidu made its Apollo 1.0 self-driving car software available as open source on Github, using the Apache/BSD license. By Day 4 of the release, it was the most downloaded C++ software on the site.

At an Apollo meetup hosted by Baidu at its Sunnyvale, California, offices, company president Ya-Qin Zhang announced Apollo 1.5, a major iteration of the software, just three months after the initial release.

"Tesla, Apple, Waymo, essentially everyone is building their own platform, their own technology. So at the beginning of the year we contemplated our own strategy. We looked at the history of PCs and mobile, and we believe an open system is more powerful, more vibrant, in the longer term. So we decided to open Apollo, both the IP, the technology and the source code."

The meetup was one way that Baidu is promoting the open-source nature of Apollo. Situated near US tech companies such as Google and Yahoo, the Sunnyvale location made it possible for engineers to walk or bike over after work. The local streets teem with self-driving research cars honing their real-world driving capabilities.

Baidu has been testing its open-source Apollo software with its own self-driving car fleet in China.

Baidu

Baidu is hardly alone in developing self-driving car technology. A multitude of companies, from automakers to automotive technology suppliers to large tech companies, have all been working on the problem. The technology has the potential to reduce or eliminate the more than 1 million deaths caused by cars around the world each year.

For the Apollo 1.5 release, Zhang said, "1.5 is a lot more capable, it has all the perception features, the sensors like radar and lidar. It has end-to-end learning technology. It has high-definition mapping access."

Sensor and high-definition mapping access are two of the biggest updates to the software. The current technology trend for self-driving cars is to load them with high-definition maps of specific roadways. These maps not only show the location of the roads, but also include fixed objects, such as infrastructure, around those roads.

The self-driving car compares what its sensors detect with its stored maps to determine its exact location. The maps also show the car where it can safely drive, and include trajectories to take when making turns in intersections. The car uses its onboard sensors to make sure it can follow the trajectory given to it on the map without hitting another car, pedestrian or anything else in the environment not depicted on the map.

These updates show that Apollo is behind the curve somewhat compared with systems being tested by other companies. Google's Waymo, for example, has been using lidar sensors with its self-driving software for years. However, Baidu's open-source approach may help it catch up much more quickly.

Open source could also help Baidu in the Chinese automotive marketplace. Zhang said that the country has over 200 automobile brands, making it difficult to sign partnership deals to develop the technology with so many. With open source, any of those companies can download Apollo and begin to develop a self-driving car on its own.

At its meetup, Baidu displayed a couple of its US research vehicles, equipped with a sensor array on the roof.

Wayne Cunningham/Roadshow

Baidu has partnered with a few Chinese automakers on its self-driving car technology. Zhang said, "We announced a partnership with Beijing Automotive, one of the biggest automakers in China, that will make Level 3 cars by 2019 and Level 4 cars by 2021." In autonomous vehicle engineering lingo, Level 3 means cars can handle some driving tasks, but must share driving with a human. Level 4 cars can handle all driving tasks by themselves, although they still have controls that let a human take over.

Open-sourcing Apollo wouldn't seem to do much for Baidu's bottom line. However, Zhang said, "Our business model in China is very clear. We are going to charge for services, such as car services, mapping services, and for some of the technology we provide, like driving simulation."

Baidu is a major purveyor of digital maps in China, and has been developing its own high-definition maps in the country. Zhang said that it has already created high-definition maps of over 300,000 kilometers (186,411 miles) of Chinese highways.

With high-definition maps playing such an important role for self-driving technology, any company that wants to develop for China will have to work with Baidu. Open-sourcing Apollo will make that easier.