Uber launches an ‘AV Labs’ division to gather driving data for robotaxi partners

Uber is making a bold move in the realm of autonomous vehicles by establishing a new division known as Uber AV Labs. This initiative aims to provide essential driving data to its more than 20 autonomous vehicle partners, including notable names like Waymo, Waabi, and Lucid Motors. Despite the initiative's focus on data collection, it's important to note that Uber is not returning to the development of its own autonomous vehicles. The company halted these efforts after a tragic incident in 2018 involving a pedestrian fatality, ultimately selling its self-driving unit in a complex deal in 2020. Now, Uber plans to deploy its own equipped vehicles across cities to gather valuable real-world data for its partners, although formal contracts have yet to be established. As the landscape of self-driving technology evolves, there is a noticeable shift from rule-based systems to those relying on reinforcement learning. This transition has made real-world driving data increasingly valuable for training autonomous systems. Uber revealed to TechCrunch that the most interested partners in this data are those already engaged in extensive data collection themselves, highlighting the challenges of navigating extreme edge cases in real-world scenarios. Currently, the capacity of an autonomous vehicle fleet limits the data that can be collected. While simulations are used to prepare for unpredictable situations, there is no substitute for extensive real-world driving experience. For instance, Waymo, which has been testing autonomous vehicles for a decade, recently faced issues with its robotaxis illegally passing stopped school buses. Access to a larger data pool could potentially help tackle such challenges before they become significant problems, as noted by Uber’s CTO, Praveen Neppalli Naga. Interestingly, Uber does not plan to charge its partners for this data at the outset. The company's primary goal is to democratize access to this valuable resource. Uber's VP of engineering, Danny Guo, emphasized the importance of laying a solid data foundation before determining the product-market fit, believing that Uber's role is crucial in advancing the entire autonomous vehicle ecosystem. The AV Labs division is in its infancy, starting with a single Hyundai Ioniq 5, which is currently being fitted with sensors like lidars and radars. Guo humorously acknowledged the scrappy nature of their efforts, noting the gradual scaling of their operation. Eventually, the fleet will be expanded, but for now, the focus remains on developing a prototype. Partners will not receive raw data but processed information tailored to their needs. Once the AV Labs fleet is operational, the data will undergo refinement to facilitate integration with partner technologies. This process will enable a deeper understanding of driving behaviors, enhancing real-time path planning. Furthermore, Uber plans to implement a 'shadow mode' where partner software will be tested alongside Uber AV Labs vehicles. Any discrepancies between the two systems will be flagged to assist in improving the driving algorithms. While Uber's scale may not match that of Tesla, which benefits from millions of customer vehicles globally, Guo remains optimistic about conducting targeted data collection based on partners' specific requirements. Looking ahead, Uber anticipates expanding the AV Labs division to several hundred employees within the next year, with ambitions to leverage its entire fleet in the future for even broader data collection efforts. The overarching sentiment from partners is clear: they are eager for any data that can enhance their autonomous driving technologies, and Uber's capacity to collect significant amounts of driving data positions them as a key player in this evolving industry.