We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 - 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!
Beijing-based Baidu last week quietly released the latest version of Apollo, its open source full-stack software solution for driverless vehicles. Apollo 5.0 — which follows on the heels of Apollo Lite, the company’s vision-based framework that enables nearly fully autonomous driving under select conditions — introduces a number of new features, including an upgraded perception model for “complex” road scenarios and a bespoke sensor calibration service.
“Apollo 5.0 is an effort to support volume production for geofenced autonomous driving,” Baidu noted in the accompanying changelog on GitHub. “The car now has 360-degree visibility … making [it] more secure and aware.”
Apollo 5.0 ushers in a brand-new data pipeline service with per-vehicle calibration options, along with spruced-up prediction evaluators and map data verification tools. Perhaps the highlights are Open Space Planner — a new scenario-based planning algorithm — and improved support for parking and for intersections, including those with stop signs and traffic lights and those without signage. Lastly, Dreamland, Apollo’s web-based simulation platform for model validation and testing, has been updated with a more robust scenario editor and control-in-loop simulation.
Baidu’s Apollo has come a long way in the roughly two years since its launch.
At CES 2018, Baidu unveiled Apollo’s second major iteration, Apollo 2.0, which introduced new reference vehicles, an encrypted framework for over-the-air updates, improved computer vision algorithms, and a system that can better determine where a vehicle is on the road. In April 2018, the company took the wraps off Apollo 2.5, which implemented improved vision-based perception, real-time relative mapping, new driving scenarios, and visual debugging tools.
Coinciding with the rollout of Apollo 2.5, Baidu launched an automotive security lab in partnership with China Automotive Technology and Research Center and China Academy of Information and Communications.
Apollo 3.0 added support for valet parking, autonomous mini buses, and autonomous microcars, as well as integration with Baidu’s voice-activated telematics software, which can perform facial recognition and monitor drivers for signs of fatigue. Moreover, it marked the incorporation of Intel subsidiary Mobileye’s Responsibility-Sensitive Safety (RSS) — a “common sense” approach to on-the-road decision-making that codifies good driving habits, like maintaining a safe following distance and giving other cars the right of way — into Apollo’s codebase.
Apollo 3.5 introduced the ability to complete unprotected turns (a notoriously challenging maneuver for driverless cars); manage speed bumps; and clear zones, side passes, narrow lanes, and parking. Additionally, it marked the launch of Apollo Enterprise, a suite of autonomous and connected services for mass-produced cars, and the Apollo Cyber RT framework, a high-performance runtime that’s compatible with lidar sensors such as Velodyne’s VLS-128.
Baidu’s nimbleness has helped it stay abreast of competition like Nvidia, which earlier this year unveiled a new full-stack automated driving solution in Drive AutoPilot. A newer challenger is Japan-based Tier IV, a University of Tokyo spinout that recently raised $28 million to develop Autoware, an open source software platform for driverless cars.
Apollo — which has grown considerably to 400,000 lines of code, or more than double the 165,000 lines of code the company announced in January 2018 — is now being tested, contributed to, or deployed by Intel, Nvidia, NXP, and over 130 global partners. (That’s an uptick from 116 partners in July 2018.) According to Baidu, the number of developers who’ve sourced Apollo’s code from the project’s GitHub repository stands at 12,000, a 20% increase from mid-2018.
Among the growing body of collaborators is California-based Udelv, which in January said it would deploy up to 100 autonomous delivery vehicles developed on Apollo 3.5 to U.S. cities, including the San Francisco Bay Area, in 2019. Other Apollo adoptees include Volvo and Ford, both of which have committed to testing Apollo-powered self-driving vehicles on Chinese roads in 2019.
Baidu is also working with Chinese automobile manufacturers Chery, BYD Auto, and Great Wall, in addition to Hyundai Kia, Ford, and VM Motori, to roll out Apollo Enterprise solutions to cars. FAW Group, which develops the Hongqi line of luxury cars, is another close partner — last year it announced plans to launch a “limited number” of Apollo vehicles across China in the following year, and Baidu says more than 60 of the world’s leading automotive manufacturers use DuerOS for Apollo — a set of AI-based IoV solutions with voice assistant, augmented reality, and motion detection capabilities — in more than 300 car models.
Baidu intends to achieve “full autonomy” on highways and urban roads by 2020, but it has competition in Beijing-based Pony.ai, which has raised $214 million in venture capital to date and in early April launched a driverless taxi pilot in Guangzhou. Meanwhile, Alphabet’s Waymo says it’s now servicing over 1,000 riders with a fleet of more than 600 cars. And GM’s Cruise Automation has been testing an autonomous taxi service for employees in San Francisco and plans to launch a public service this year. Other rivals include Tesla, Zoox, Aptiv, May Mobility, Pronto.ai, Aurora, and Nuro.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.