With this year’s global autonomous driving technology experiencing a development bottleneck and going through Argo.ai’s impactful closure, its commercialization seemed nowhere in sight. However, China’s technological pioneer, Baidu, announced its plans to build the world's largest autonomous ride-hailing service, Robotaxi, in 2023 at Apollo Day, the autonomous driving platform expo, at the end of November. Since Baidu’s Robotaxi launch in Chongqing and Wuhan this August, the service has established operations in ten major cities. In Q3 alone, Apollo Go (aka “蘿蔔快跑” in the Chinese market), a taxi service platform, processed 470k trips, 65% more than the last quarter. Take Beijing and Shanghai as examples, each Robotaxi in the city can complete fifteen trips on a daily basis. Up to the end of Q3, Apollo Go had provided 1.4 million taxi services. Due to Baidu’s many years of autonomous driving technology developments and big data analysis, Robotaxi’s operation can now expand from designated areas to public roads. While only twenty days are needed to deploy autonomous driving services to a new city as claimed by Apollo Go, we can expect more cities to embrace its Robotaxi service next year.
An unresolved critical aspect of autonomous driving tech’s realization and commercialization was so called corner cases (aka long-tail problems) that haven’t been experienced by the systems. However, Baidu recently announced the first AI big model WenXin in the autonomous driving industry. The “pre-trained visual language model with weak supervision” model can recognize thousands of target objects, thanks to WenXin’s billion-parameter scale, greatly expanding the scope of semantic recognition, such as special vehicles, plastic bags, etc. By training small models with larger ones, autonomous driving sensors’ generalization potential can rise significantly and be able to tackle even corner cases. Additionally, Baidu’s HD map boasts a 96% autonomous data gathering for setup, utilizing hundreds of millions of Chinese drivers’ experience and expertise to build a 12 million km road network database based on time and space. The digital map consists of multiple information layers, including stationary, mobile, knowledge, and driving, and can effectively establish driving knowledge graphs, safety driving, and comfortable driving behavior patterns. In the future, Baidu will simultaneously push L4 and L2+ tech while they complement each other: L4 will input new functions to L2+, and L2+’s feedback after its popularization can elevate L4’s generalization potential in turn.
Additionally, Baidu’s self-developed Kunlun AI 2nd-gen 7mm chip possesses end-to-end autonomous driving computing power, further integrating software and hardware. Till now, Baidu’s autonomous vehicles developed with auto companies such as Geely, Arcrfox, Weltmeister, Aion, etc., will take on the mantle of Robotaxi’s operation and its consumer experiences with 100 new offline experience centers across 46 cities in China, commencing large-scale autonomous driving AI training and elevating its technology level. On the other hand, global autonomous driving tech pioneer Waymo and Geely’s Zeekr recently announced their concept car, M-Vision, in Los Angeles. According to my estimations, 2023 will be the ultimate showdown for Baidu Apollo and Waymo in public road testing, and autonomous driving technology’s full application schedule will fall into the hands of Baidu or Waymo.