At the recently held 2025 Yunqi Conference, Hellobike Chief Technology Officer Liu Xingliang delivered a keynote speech titled "Full-Stack AI Architecture Drives Transportation Transformation: From Travel Agent to Future-Oriented Travel AGI," systematically outlining Hellobike's strategic layout and practical achievements in artificial intelligence technology. Liu Xingliang stated that the AI revolution is profoundly reshaping the transportation industry, with Hellobike advancing through a dual-drive approach of "information intelligence + physical intelligence" to accelerate the construction of a future-oriented smart transportation system and fully promote the commercial deployment of Robotaxi services.
"AI's impact will exceed that of the Industrial Revolution and Internet Revolution," Liu Xingliang noted. As two major branches of the AI revolution, information intelligence represented by internet platforms and physical intelligence represented by embodied robots and Robotaxis are both developing rapidly. By combining information intelligence and physical intelligence, Hellobike is committed to building safer, greener, and smarter transportation solutions.
According to Liu Xingliang, leveraging various advanced AI technologies, Hellobike continuously explores how to use AI Agent products to better understand and meet user needs while enhancing user experience. For example, Hellobike's homepage situational awareness Agent can constantly monitor user needs, identify user behavior and status in real-time, and provide personalized services promptly. The Hellobike ride-sharing matching Agent serves as a "high emotional intelligence digital twin," which can assist in negotiating highway tolls, timing, and other matters under user initiative or system automatic trigger, effectively resolving transaction conflicts and helping drivers and passengers reach mutually satisfactory deals. Additionally, Hellobike has launched a "Travel Assistant" multi-Agent collaborative system centered on voice interaction, supporting various complex tasks including shopping guidance, order placement, rule queries, and POI responses, significantly shortening user decision paths.
In terms of cost efficiency, Hellobike creates digital employees for rich scenarios. In the CodeAgent field, it builds extensive MCP tools and rules combined with platforms like Tongyi Lingma to achieve end-to-end AI programmers from requirements to code. In customer service, it has implemented a managed model. In AI review, it has achieved comprehensive inspection in areas including work orders, driver-passenger recordings, and transaction evaluations.
Simultaneously, Hellobike relies on its self-developed agent construction platform supporting multiple modes including workflow, agentic, and SDK, along with self-developed hybrid inference models. Through reinforcement learning and multi-step training, it has significantly reduced online inference time, making inference models available for C-end users while substantially reducing knowledge base complexity. In computing power, it has achieved multi-cloud, multi-card management, unified scheduling, and training-inference acceleration, significantly improving resource utilization.
Behind these AI capabilities is Hellobike's deep application of data mining and large model technologies. Combined with Alibaba Cloud's full-stack AI architecture, Hellobike's AI platform has achieved comprehensive architectural upgrades, laying the foundation for both parties to jointly advance Robotaxi technology research and application.
Liu Xingliang pointed out that in the Robotaxi field, Hellobike is building a "vehicle + cloud collaborative" intelligent driving super brain, committed to achieving "fast, accurate, and stable" decision-making. Based on this, Hellobike has created its independently developed "end-to-end L4 technology system," with its first pre-installed mass production Robotaxi model "HR1" (Hello Robot1) making its global debut at the 2025 Bund Conference.
To support this advanced architecture, Hellobike has constructed a three-in-one technical foundation of "big data, big computing power, and big models." In terms of big data, through automated compliance production lines, Hellobike plans to build a high-quality dataset containing no less than 10 million clips by 2026. Meanwhile, leveraging deep cooperation with Alibaba Cloud, Hellobike will build a ten-thousand-card-level computing power cluster for training and iterating autonomous driving models, and develop the "Daoyu" hundred-billion-parameter large model, constructing three major cloud capabilities including data mining, World Model simulation, and VLA cloud driver foundation. Simultaneously, Hellobike is actively deploying "cockpit AI Agents," creating intelligent entities such as smart cockpit main control Agents, vehicle assistants, and travel assistants through multimodal recognition, RAG and knowledge base technologies, redefining travel spaces.
In September this year, Hellobike announced that its Robotaxi business received strategic investment from Alibaba Group, with both parties further deepening comprehensive cooperation in intelligent driving large models, computing power platforms, and Robotaxi fields to jointly accelerate the commercialization and scaling process of the Robotaxi industry. At this Yunqi Conference, Liu Xingliang stated that Hellobike will use full-stack AI technology as its engine to promote deep integration of information intelligence and physical intelligence, providing users with greener, safer, more convenient, and affordable smart transportation services.