--- title: "Former Horizon Robotics Executive Launches Startup for Foundational Robot Intelligence, Secures Tens of Millions in Seed Funding" type: "News" locale: "en" url: "https://longbridge.com/en/news/286870213.md" description: "The company was established a few months ago" datetime: "2026-05-19T07:18:44.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/286870213.md) - [en](https://longbridge.com/en/news/286870213.md) - [zh-HK](https://longbridge.com/zh-HK/news/286870213.md) --- # Former Horizon Robotics Executive Launches Startup for Foundational Robot Intelligence, Secures Tens of Millions in Seed Funding News on May 19: "Dingdang Power," a startup specializing in large spatial intelligence models, announced the completion of a seed financing round worth tens of millions of yuan. This round was led by HORIZONROBOT-W, with Zhengjing Fund participating as a co-investor. It is understood that the funds will be primarily allocated to the development and validation of the "Large Spatial Intelligence Model + Physical Agent" framework, as well as the establishment of a closed-loop system for real-world scenario data. At a time when investment in embodied AI is becoming increasingly rational, this financing reflects industrial capital's phased preference for teams with comprehensive experience ranging from foundational algorithms to hardware deployment. Public information shows that Dingdang Power was established only a few months ago, with its core leader being Niu Jianwei, a former executive at Horizon Robotics. A review of his career history reveals an industry technology veteran who grew alongside the wave of deep learning in China. Niu Jianwei previously worked in the Voice Technology Department at Baidu. In 2012, he participated in early GPU-based deep learning model training at Baidu IDL (Institute of Deep Learning). In 2015, invited by Yu Kai, Niu joined Horizon Robotics during its startup phase, serving in roles ranging from Algorithm Engineer to General Manager of the Intelligent Cockpit product line. This complete career trajectory, spanning algorithm construction, chip adaptation, and final product mass production, is a relatively scarce background in the current landscape of large model startups. Most AI entrepreneurs in the past have tended to focus on the algorithm layer. However, in the real physical world, the deployment of AI is severely constrained by computing platforms, sensor precision, and systems engineering capabilities. Previously, in 2023, Niu Jianwei proposed the concept of low-cost data stimulation for vertical domain Post-training, demonstrating his long-term focus on the core issue of "cost and scale" in AI commercialization. This also served as an important precursor to his subsequent decision to enter the embodied AI track from the underlying architecture. According to disclosures, Dingdang Power's core business focuses on the "Large Spatial Intelligence Model + Physical Agent." To understand the essence of this business, it must be observed within the context of the current evolution of robotics technology. Over the past few decades, the traditional robotics industry has mainly relied on cybernetic thinking, i.e., writing rules and adjusting parameters for single tasks, resulting in very weak scenario generalization capabilities and high customization costs. With the explosion of large models, the industry began attempting to bundle vision, language, and action for training, forming VLA (Vision-Language-Action) models. However, current VLA models are facing obvious practical shortcomings in engineering: significant difficulty in matching multimodal data, limited model scalability, and the challenge of achieving low-cost continuous self-evolution in the physical world. Dingdang Power's technical route attempts to avoid this blind spot. Its solution is not merely fine-tuning a single algorithm but aims to build a system-level solution: letting the large spatial intelligence model handle the understanding of complex physical environments, while the Physical Agent serves as the execution layer, responsible for deep integration with physical entities. The essence of this architecture is an attempt to bridge the engineering gap preventing general large models from outputting intelligence from the "digital screen" to "physical entities." In the current capital environment, financing in the large model sector is concentrating on targets with clear business models and engineering deployment capabilities. The fact that Dingdang Power secured leading investment from Horizon Robotics in its early stages can be attributed to two core logics: First, deep strategic synergy value. As a provider of underlying intelligent computing platforms, Horizon Robotics' core strategy is not limited to intelligent driving but extends its software-hardware integration capabilities to a broader ecosystem of robotics and embodied AI. Dingdang Power's exploration at the Physical Agent level can be seen as a natural extension of Horizon's computing ecosystem into physical entities. Second, the capability to build a closed-loop system for mass production data. This financing round explicitly stated that funds would be used for "building a closed-loop system for real-world scenario data." In the field of embodied AI, the core barrier lies in who can acquire the highest quality physical world interaction data at the lowest cost. Niu Jianwei's practical experience in the intelligent cockpit sector has given him a deep understanding of the flywheel effect of mass production data feedback and model iteration. Compared to purely academic teams, this systematic capability to establish an industrial-grade closed loop for data cleaning, labeling, and training is a fundamental asset valued more highly by industrial capital. Looking ahead, the main battlefield of AI is shifting from content generation in purely digital dimensions to entity intelligence with strong interaction with the physical world. The "spatial intelligence" track where Dingdang Power operates is precisely the core infrastructure during this transition period. However, this remains a long and challenging track full of unknowns. Long-tail scenarios in the physical world are extremely complex, and physical common sense such as lighting, friction, and gravity still represents a huge cognitive gap for current large models. For Dingdang Power to complete the construction of a high-quality data closed loop in the short term, it will inevitably face huge data collection costs and engineering challenges in adapting to multi-terminal hardware. The key to observing the future development of this enterprise lies in whether it can deliver benchmark cases of physical agents with certain generalization capabilities and controllable costs in real commercial scenarios within one to two years. ### Related Stocks - [09660.HK](https://longbridge.com/en/quote/09660.HK.md) - [BIDU.US](https://longbridge.com/en/quote/BIDU.US.md) - [09888.HK](https://longbridge.com/en/quote/09888.HK.md) - [89888.HK](https://longbridge.com/en/quote/89888.HK.md) ## Related News & Research - [The Seed 100: The best early-stage investors of 2026](https://longbridge.com/en/news/286743225.md) - [Figure AI had one of its robots race a human to sort packages. It lost.](https://longbridge.com/en/news/286883540.md) - [The Seed 40: The best women early-stage investors of 2026](https://longbridge.com/en/news/286743171.md) - [ElliQ is a surprisingly helpful companion robot for older adults](https://longbridge.com/en/news/286673621.md) - [Unitree Robotics and DEEP Robotics Accelerate IPO Progress; Shoucheng Holdings' Robotics Investments Enter a Value Validation Phase](https://longbridge.com/en/news/286853966.md)