Pudong's embodied artificial intelligence sector has recently witnessed several significant advancements. Zhiyuan has unveiled the world's first end-to-end motion foundation model, fundamentally redefining the control capabilities of humanoid robots from the ground up in terms of motion intelligence. Meanwhile, Qingtianzu, after announcing the completion of a multi-billion yuan financing round, has swiftly entered into a comprehensive strategic partnership with JD.com to accelerate the large-scale commercial deployment of robots. With core technological breakthroughs in humanoid robotics and commercial application scenarios advancing synergistically, Pudong is accelerating the establishment of a scaled pathway from "R&D to service to application," aiming to build a highland for the embodied AI industry.
**Zhiyuan Unveils World's First End-to-End Motion Foundation Model, Comprehensively Upgrading Underlying Motion Intelligence**
Zhiyuan has officially launched the world's first end-to-end BFM-2 motion foundation model, marking a comprehensive upgrade to the underlying motion intelligence of humanoid robots. The newly released BFM-2 model is the first large-scale full-body motion foundation to incorporate an end-to-end generative training mechanism. By performing continuous probabilistic modeling of the robot's full-body dynamic state space, it has created the world's first motion foundation model capable of recovering from "any current state to any random target configuration." This model breaks through the previous bottleneck of "pre-set trajectories and passive execution" in humanoid robot motion control. With its novel generative motion architecture, it endows humanoid robots with human-like motion reasoning capabilities and full-state autonomous generation abilities, redefining the technological benchmark for embodied AI motion control.
Unlike traditional robots that replicate teleoperated actions one-to-one and rely on manually designed transitional trajectories, the BFM-2 motion foundation model possesses genuine motion reasoning capabilities. Relying on its unique two-stage architecture, it first autonomously recognizes its real-time dynamic state and parses high-level task intent, then autonomously deduces the complete evolution path from the current state to the target state, achieving a revolution from "action replication" to "motion reasoning."
Furthermore, unlike traditional robot motion control systems that only have limited state fall-recovery capabilities, the BFM-2 motion foundation model is the first to achieve generative motion deduction and dynamic configuration re-planning for the full state space. The industry believes the BFM-2 model is the first to give robots a motion "cerebellum" foundation with truly generalized generative capabilities.
It is reported that BFM-2 will be further upgraded to BFM-3, expanding the motion foundation model's input from pure proprioceptive state to incorporate higher-dimensional multimodal states including vision, touch, speech, spatial semantics, and environmental topology. This evolution will transition from a "generative generalized motion foundation" to a "multimodal motion foundation," truly constructing an embodied motion center that integrates spatial perception and multimodal understanding.
**Qingtianzu Completes Multi-Billion Yuan Financing and Partners with JD.com to Accelerate Large-Scale Robot Application**
Qingtianzu recently announced the completion of its Series A and A+ financing rounds, raising several billion yuan. Following this round, Qingtianzu's valuation has reached 7 billion yuan, officially placing it among unicorn companies. Qingtianzu stated that the funds will be primarily used for building its RaaS (Robot as a Service) system and expanding into industrial and other scenarios, aiming to transition robots from "being used at events" to real-world settings closer to production and operations, such as industrial manufacturing, park management, warehousing and logistics, and commercial services.
Li Yiyan, CEO of Qingtianzu, revealed that the challenge in deploying robots into real scenarios is not merely "whether robots are available," but whether stable, replicable application delivery capabilities can be established. "Especially in industrial and park scenarios, clients are no longer focused on short-term demonstration effects, but on whether robots can operate continuously, be dispatched on demand, maintained promptly, and create tangible value in specific tasks."
Having been established for less than six months, Qingtianzu has already built its RaaS application service platform, successively launching core systems and functional modules including partner SaaS, a global app, intelligent dispatch, and a skill store. These provide foundational support for cross-regional robot scheduling, standardized delivery, and multi-scenario operations. To date, the Qingtianzu platform can schedule over 4,000 robots, with its service network covering multiple cities across China. It has accumulated extensive delivery experience in scenarios such as cultural tourism, commercial performances, exhibitions, retail, office environments, and education.
Subsequently, Qingtianzu announced a comprehensive strategic partnership with JD.com. The two parties will engage in deep cooperation around co-building product solutions, empowering channel supply chains, and large-scale procurement. In the core leasing segment, Qingtianzu will open an official flagship store on JD.com, further expanding the online entry point for robot leasing services. They will collaborate on supply chain coordination, online transactions, offline delivery, and operational services, promoting the evolution of robot leasing from single-point services towards more standardized, platform-based application delivery.
Addressing industry client needs, Qingtianzu and JD.com will continue to collaborate on product co-development, focusing on diverse scenarios like education, guided tours, and cleaning to jointly create customized solutions and service ecosystems better suited to real application requirements. At the channel and procurement level, the partners will combine JD.com's online traffic platforms with offline formats like JD Malls to implement application scenarios, and promote large-scale procurement and scenario-based application deployment in suitable settings.
Qingtianzu stated that this strategic partnership is a significant attempt to deepen the RaaS model. In the future, it will continue to launch customized solutions for different clients and scenarios, facilitating the large-scale deployment and popularization of the robotics industry.
Comments