
More Than a Chip: Why Horizon Robotics' Full-Vehicle Agent Won a World-Class Technology Award
More Than a Chip: Why Horizon Robotics' Full-Vehicle Agent Won a World-Class Technology Award
The Wuzhen World Internet Conference announced this year's Leading Technology Awards. Horizon Robotics won for its full-vehicle agent technology platform, sharing the list with international projects such as Microsoft Copilot, AMD, and Alibaba Cloud Qoder. Seeing this list, a natural question arises: Horizon Robotics is known for its intelligent driving chips, so why was this award given to an entire platform rather than just a single chip?
Understanding this award as just another honor for the Journey chip would underestimate its significance. This award does not recognize the compute metrics of a single chip, but rather Horizon Robotics' comprehensive capability to bring intelligent driving from the laboratory to real roads and achieve large-scale mass-production deployment—in other words, a complete physical AI system. Currently, the focus of large-model competition remains on text and image generation, while the difficulty of physical AI was never about parameter scale: it must drive stably under real road conditions such as backlighting, sudden pedestrian emergence, and worn lane markings, while meeting functional safety and regulatory requirements throughout. The automobile is precisely the first scenario where physical AI achieves large-scale deployment.

Horizon Robotics' "Full-Vehicle Agent Technology Platform for Physical AI" won the 2026 World Internet Conference Leading Technology Award. Image source: Horizon Robotics
Horizon Robotics' full-vehicle agent technology platform consists of four layers: the underlying Journey BPU intelligent driving chip, the cockpit-driving integrated Starry Sky chip, the full-vehicle agent operating system Kaka Shrimp, and the top-layer HSD end-to-end assisted driving system. In traditional electronic and electrical architectures, the cockpit and intelligent driving are independent of each other—one handles audio-visual and voice interaction, the other handles perception and decision-making, and the two exchange information through a gateway. Although this isolated design reduces the risk of mutual coupling, it also brings additional hardware costs, and cross-domain communication has latency, making it difficult to form unified full-vehicle intelligence in complex scenarios.

At the 2026 World Internet Conference, Horizon Robotics introduces its full-vehicle agent technology platform (screen displays "Algorithms Define Hardware · Hardware Feeds Back Algorithms"). Image source: CCTV.com
The key capability of the Starry Sky chip is integrating the cockpit and intelligent driving onto the same chip. This is not simply stacking two functions together, but a reallocation of the vehicle's computing power: in scenarios requiring high perception loads, such as intersection cut-ins and low visibility in rain, computing power prioritizes the perception and decision-making of intelligent driving; under relatively simple conditions such as highway cruising, redundant computing power is scheduled to the cockpit-side large model to support voice and interaction services. This mechanism of dynamically allocating computing power according to scenarios was difficult to achieve with the previous separated cockpit-driving architecture.
HSD adopts an end-to-end technical approach and is China's first mass-produced one-stage end-to-end assisted driving system, achieving ultra-low latency from photon input to trajectory output. On the technical route of intelligent driving, the choices of various companies differ markedly: XPeng bets on a general world model with VLA, Huawei ADS focuses on a mapless urban solution, and Li Auto emphasizes stable experience in family scenarios. Horizon Robotics chooses an open end-to-end route that can adapt to multiple automakers and multiple vehicle models—it does not build cars, nor does it bind its technology to its own vehicle models. Instead, it delivers chips, operating systems, and toolchains as a whole, and automakers define and tune their products on this basis. In other words, Horizon Robotics positions itself as an enabler at the industry's foundation rather than personally building a closed ecosystem.
Horizon Robotics · Open Supply Chain | Huawei / XPeng / Tesla · Closed-Loop In-House Development |
|---|---|
Positioning: Industry foundational infrastructure, does not build vehicles | Positioning: In-house algorithms bound to own vehicle models |
Physix Frontier