In August, at the 2026 World Robot Conference (WRC), Zhiyue Spatial Intelligence officially unveiled its next-generation brain-inspired simulation platform, the "Zhiyue Lingwu (DeepSoma)". According to the company, this is a whole-brain simulation foundation platform designed for brain science and brain-inspired intelligence research, offering integrated modeling, training, solving, and simulation capabilities for the Brain, Environment, and Body. The platform's launch video showcased the company's self-developed "digital fruit fly," demonstrating the entire process from reconstructing a real whole-brain connectome to autonomous perception, decision-making, movement, environmental interaction, and cross-embodiment transfer. The company positions this platform as a foundational tool for whole-brain simulation, aligned with the national "15th Five-Year Plan" frontier technology research directions. With a philosophy of "starting from the real brain to find the source code for next-generation intelligence," the company advocates for AI to transition from "data-driven" to "structure-driven" approaches.

On the capital front, Zhiyue Spatial Intelligence has also been highly active: completing five consecutive funding rounds between February and July 2026, with the team stating that a final round will close in September. In the view of founder Qian Linrui, current AI large models driven by massive data and immense computing power are approaching their limits. Internet data is being fully consumed, and single-modal data cannot support AI in truly entering the physical world. In contrast, the human brain, operating on just about 20 watts of power, can accomplish complex perception, decision-making, and execution. Therefore, Zhiyue has chosen an "inverse" path: starting from a 1:1 replication of a biological brain to discover a next-generation model foundation that does not rely on massive data or enormous computational resources. At the WRC, we had the opportunity to speak with Dr. Qian Linrui about this "intelligent brain" roadmap. The following is an edited excerpt of that conversation.

Where to begin with the platform

Regarding the "intelligent brain" concept that Zhiyue champions, Dr. Qian explained the key highlights and motivation behind the new product. He noted that current AI model foundations are dominated by the large model paradigm, which is already showing its limits because internet data has been exhausted. These models are purely data-driven, meaning their ceiling in intelligence is visible. Another issue is data modality. The one-dimensional and two-dimensional image data, as well as semantic text data from the internet era, are essentially single-modal. However, for AI to move from the big screen into real physical spaces, it requires multimodal data. The question of how to integrate multimodal data and how to build AI models for it remains a significant challenge. Many teams are exploring concepts like "latent space" or "world models" to find a feature space where different modalities can be aligned. The brain itself already possesses this multimodal processing capability, so Zhiyue aims to borrow from the brain's architecture to find solutions. A third critical factor is the growing importance of low power consumption, low computational demands, and autonomous learning in the Physical AI era, all of which are forms of intelligence directly learnable from the brain. The human ability to generalize from one example to many, or to derive a hundred insights from a single case, while completing complex tasks with minimal energy, is highly valuable. This led the team to consider whether replicating a biological brain could pave the way for a new model framework and foundation that is not driven by big data or massive computing in the next era.

Compatibility across different robot forms

When asked about the platform's potential for universal application across various embodied robots, Dr. Qian affirmed that it is compatible with all forms. Zhiyue is building a model foundation, and in the Physical AI era, embodiment is just one segment, albeit one where the company is progressing rapidly. When they started in 2025, they rode the wave of embodied AI, though many institutions at the time considered their technical route non-consensus, believing that "AI large models plus embodied data would surely train a good model." However, from 2025 to now, despite significant capital and data influx, embodied robots have still seen slow industrial adoption, largely stuck at the Sim2Real gap, delivery cycles, and delivery costs. Looking back, the team is firmly convinced that traditional AI approaches cannot solve this. So, what they are doing is directly placing biological models onto robots of various forms—single-arm, dual-arm, wheeled, or humanoid—and letting them perform real tasks in specific scenarios, which clearly demonstrates the value of their approach.

Selecting application scenarios

On the relationship between Zhiyue's technology direction and its various application scenarios, Dr. Qian outlined two main considerations. First is the demonstrative effect. In terms of industry selection, they have chosen a focused approach, targeting one breakthrough scenario each in the B2B sector and the broader consumer (pan-C-end) market. Second is the technical route. Current robot delivery methods vary, as many companies are flush with funding and not operating on purely commercial logic; some are willing to operate at a loss to capture market share. Zhiyue prefers not to compete in that price-war arena. When choosing scenarios, they consider whether the task is something that cannot be accomplished just by throwing money at it—if it lacks new technological breakthroughs, it's likely a project more worthwhile for them to invest in. Their first project, a deep-sea robot operating tens of meters underwater without any ambient light, presents a difficulty far greater than ground-based or high-altitude robots. In large consumer and large enterprise scenarios, they anchor on areas where their technology provides an absolute advantage, including cost and performance, aiming for replicability under pure commercial logic in the future.

Future directions and long-term vision

Looking ahead, besides embodied scenarios within the Physical AI space, Zhiyue aims to advance biological scenarios—including virtual cells and brain-computer interfaces—with a focus on providing models for these industries. A longer-term goal includes "digital immortality." Regarding short-term and long-term objectives, Dr. Qian outlined that within 1-3 years, they hope to complete a vertebrate brain that approaches 70% to 80% of the human brain's complexity. In about five years, they aim to replicate a complete adult human brain, transfer it to robots and computers, and use this digital brain to investigate whether consciousness is a product of the biological brain, body, and environment interaction, or if it could also emerge from complex computational structures. They want to explore whether a machine can discuss its own emotions, self, and experiences, and whether we can establish testable, falsifiable scientific criteria for consciousness that go beyond intuition. They also question whether all high-level and low-level intelligence resides entirely within the brain's connection structure. If the brain can be replicated, could it potentially slow brain aging or form a new sense of self? Ultimately, true digital immortality is the plan for the coming years.