We Connected Digital Agents to a Physical Robot at WRC 2026 — Why Robots Need Two Brains

At the World Robot Conference (WRC) in August, a robot joined the podium during the award ceremony, showcasing a significant development. This robot was part of a joint project between Mininglamp and HIKROBOT, and it sparked a keynote by Mininglamp founder Minghui Wu. Wu made a bold statement: for robots to become part of commercial production systems, they need "two brains." The focus of this article is not on the robot or the keynote, but rather on the meaning behind Wu's judgment.

The "first brain" in the robot is its ability to process visual input and convert it into physical actions. This is the core of what many researchers in embodied AI are working on, with models like OpenVLA, the π0 series, and VLA-JEPA leading the way. However, when these models are applied to real-world scenarios like restaurant cleaning, warehouse logistics, or patrol routes, new challenges emerge.

How do multiple robots coordinate their tasks? How do they understand and process new orders in a kitchen or a patrol route? These issues lie outside the scope of the visual input-action output models and require a separate "second brain" for orchestration.

The "second brain" is the organizational layer that manages the coordination of multiple robots, the dispatch of tasks across systems, and the connection to a company's existing IT infrastructure. This is not a new problem for Mininglamp, which introduced the HAO framework in 2018 to connect humans, digital agents, and physical robots into one network.

This framework aimed to generate productivity through division of labor, rather than expecting any single entity to be fully capable on its own. The joint WRC booth demonstrated how this orchestration capability could extend from digital worlds into the physical world.

Interestingly, the technological advancements in digital agents and physical robots are following parallel paths. Both have moved from hard-coded systems to more autonomous, end-to-end models. They also share similar memory architectures, with both requiring spatial memory, task trajectory memory, semantic memory, and skill memory.

However, there is one crucial aspect that needs bridging: the ontology of embodied contexts. While physical structures are typically referred to as ontology in embodied contexts, semantic memory deals with abstractions of concepts and relationships. For a robot to truly understand and communicate with humans or other agents, a shared semantic layer is necessary.

To make this organizational brain function, a communication protocol is needed, akin to email. This protocol should be open, simple, and vendor-neutral, allowing robots, digital agents, and legacy systems to communicate effectively. The digital world has made progress in this area, with the development of MCP and A2A protocols. As robots join this landscape, the protocol must accommodate more participants and address the latency requirements for real-time interactions.

In essence, the pursuit of "two brains" for robots is not a new field, but rather an extension of the continuous evolution of digital agents.