The Rise of Context Engineering: The New Hotspot After Prompt Engineering and RAG}

"Huawei offers an alternative solution for embodied intelligence."

At HDC 2025, Huawei launched the CloudRobo embodied intelligence platform, serving as the "technical foundation" for embodied intelligence. It leverages cloud-based "strong intelligence" to empower physical robots, bypassing the slow and costly deployment of onboard AI, and exploring the broadest and fastest path to practical embodied intelligence.

"Huawei Cloud's goal is to make all connected entities into embodied intelligent robots," said Zhang Pingan, CEO of Huawei Cloud Computing.

By shifting focus from creating "embodied forms" to providing cloud-based technological empowerment, Huawei’s strategic approach offers a new perspective on the development of embodied intelligence.

Embodied intelligence is not about the "morphology" or the level of onboard intelligence but about the ultimate goal of "more practical" use—ranging from humanoid robots to mobile robots and trucks—accelerating their deployment in the physical world.

This end-goal thinking greatly broadens the imagination for industrialization and points to the most efficient pathways for commercial implementation.

Practical applications in industry confirm the feasibility: in spray painting, CloudRobo helps ET robotic arms quickly adapt to new tasks; in semiconductor manufacturing, CloudRobo enables logistics robots to synchronize with production systems, update task plans, and handle material transport.

Partners like UAI and ET have already achieved large-scale commercial applications. Robots seamlessly navigate factories and perform massive operations. While the industry debates when humanoid robots will enter the productivity era, these robots, already widely deployed, are leading the way in realizing embodied intelligence’s productivity value in real scenarios.

Thus, a more pragmatic and clear development path for embodied intelligence has emerged: abandoning overemphasis on a single form, focusing instead on efficient, universal AI empowerment to activate the potential of existing and future machines, with productivity improvement as the benchmark, creating a scalable value loop. This marks the industry’s move toward maturity.

What the scene needs is not the "form" but productivity

Huawei Cloud uses a simple yet direct image to illustrate embodied intelligence. Besides humanoid robots in the spotlight, it also includes mobile robots in industrial scenes and collaborative arms busy on production lines. Beyond having "body" and "brain," they share another trait: productivity.

The industry’s common view is that humanoid robots represent the "ultimate form" of embodied intelligence, mainly because of their "productivity potential." With a human-like appearance and capable of similar actions, they can perform tasks within the scope of human capabilities and seamlessly integrate into human-scale physical spaces.

However, the core of this view is to pursue "broader task execution capabilities," emphasizing "productivity" over form. The key is whether robots can provide effective solutions to real problems—form is secondary.

In industrial manufacturing, highly standardized processes, mature automation, and structured environments make it the primary application area for embodied intelligence. The critical requirement here is reliability—robots must operate with high stability to meet the core goal of quality and efficiency improvement.

For example, in an 8-inch wafer factory, UAI’s OW8 wafer handling robots automate the entire process from lithography to cleaning. They use high-precision SLAM navigation to avoid obstacles and plan paths autonomously. Their unique four-sided open chassis design simplifies maintenance, reducing downtime by over 60%. They also feature patented shock absorption to keep vibrations below 0.1g, reducing wafer damage.

In daily operations, a single OW8 robot handles over 240 tasks, with the entire system processing more than 12,000 items daily, fully supporting 24/7 continuous production.

UAI’s case confirms that the essence of scene-specific "productivity" is simply being capable of doing the work.

Beyond factories, high-demand commercial environments also require robots to operate in dynamic, real-time conditions. Companies like Qinglang and Yunji are transforming from simple delivery to embodied intelligence, integrating robots into workflows for multi-task execution such as delivery and cleaning, further reducing human involvement.

From industrial to commercial scenarios, the clear development path for embodied intelligence is to create a universal "productivity engine."

Whether it’s mobile robots in semiconductor factories or delivery robots in restaurants and hotels, their value lies in reliable operation, deep integration into workflows, solving efficiency bottlenecks, freeing human labor, and generating quantifiable economic benefits.

Companies like UAI, Qinglang, and Yunji have demonstrated that diverse physical forms equipped with powerful "brains" are turning the vision of a "productivity era" into reality across various scenarios. The future of industry is not about chasing a single form but about adapting this "productivity engine" to broader scenes, driving more efficient automation, and realizing ubiquitous machine intelligence in the physical world.

The ultimate goal of embodied intelligence is the emergence and evolution of productivity tools.

Not opposition, but coexistence

After analyzing scene needs, a new question arises: what is the future of humanoid robots that are not yet widely deployed? Are they an iteration or a coexistence with multi-form embodied robots?

Here’s a relatable example: a large, thriving company with a deep-tech R&D team and a management team that efficiently connects various business lines. The two are not replacing each other but working together, complementing each other’s strengths.

In terms of embodied productivity, the "expert" robots are the current practical embodiment, performing non-standard tasks across different roles. Their collaboration with other forms creates the most suitable solutions for specific scenarios.

When diverse embodied robots work in "cluster collaboration," the new challenge is how to enable tight coordination—understanding tasks at millimeter precision and seamless communication among robots.

Currently, there are several technical routes: some focus on enhancing the intelligence of the physical body, others on iterative improvements from the base platform. For example, UAI’s MAIC system combines a multi-modal general-purpose base model + "one brain, multiple forms" architecture, integrating intelligent control with modular hardware. It endows robots with smarter control and enables group collaboration.

The multi-modal general-purpose base model handles complex planning and reasoning, trained on a multi-modal data platform, compatible with hundreds of hardware types, and supports open-source datasets. The core is a multi-modal VLM model, with added planning and evaluation modules for converting instructions into control commands, and online assessment of control quality.

The "one brain, multiple forms" control model manages real-time control of multi-form robots, centered on the MAIC (Mobile AI Comprehension) system, enabling multi-modal perception, multi-arm collaboration, multi-form mobility, and global logistics scheduling. It combines reasoning with high efficiency and precision, forming the core of high-generalization embodied robots.

UAI also built an Agent aggregation platform for various industries, integrating large models with industrial software for training, fine-tuning, and rapid adaptation to complex downstream tasks, including expert agents with reasoning capabilities and global production optimization.

Videos released by UAI show multi-configuration robots working in harmony: mobile robots perform precise actions, humanoids break down complex tasks with detailed motion adjustments, and the collective cognition of multi-form robots illuminates the "beam" of embodied intelligence.

Therefore, the ultimate competition in embodied intelligence is not between "humanoid" and "multi-form" paths but who can first build a universal, open, and efficient "group intelligence collaboration"—a comprehensive "intelligent productivity network" covering the physical world. This requires breaking free from the mindset of single-body intelligence and embracing a collaborative ecosystem focused on productivity.