Doosan Robotics to Develop Humanoids That Work Like Skilled Technicians... Proof-of-Concept in 2028

Kim Minpyo: "In the future, the primary operator of robots will be agents, not humans"

Doosan Robotics is embarking on the development of humanoid robots designed to "work like skilled technicians" on industrial sites. The company plans to collect onsite data using collaborative robots, first perfect task intelligence, and then gradually scale up robot bodies from single-arm to dual-arm and eventually to mobile platforms. By 2028, Doosan aims to conduct proof-of-concept (PoC) testing for industrial humanoids.

Kim Minpyo, CEO of Doosan Robotics, is presenting on the humanoid strategy. Photo by Paek Jongmin, Tech Specialist

Kim Minpyo, CEO of Doosan Robotics, stated at the "2026 Global Machinery Technology Forum" held at the Korea Institute of Science and Technology on September 7, "The issue is not whether we will pursue humanoids or not, but the order in which we do so," adding, "We will first define task intelligence and then build bodies that fit those needs."

The roadmap presented by Doosan Robotics consists of three stages: collaborative robots, intelligent robot solutions, and finally, industrial humanoids. The company is currently accumulating real-world operational data through collaborative robots already deployed in industrial settings. Based on this data, Doosan is developing intelligent robot solutions capable of autonomously performing skilled tasks such as surface treatment and welding. The process involves progressing from single-arm operations to dual-arm tasks, and then integrating mobility functions to ultimately expand into industrial humanoids.

Kim emphasized that physical AI should not be viewed as merely an extension of large language models (LLMs). Unlike generative AI, which can leverage the vast amount of information accumulated on the internet, robots must generate and handle real-world data such as the sequence and force of contact, friction, and the success or failure of tasks.

He explained, "While intelligence is advancing rapidly, even the smartest models cannot handle a cup like a human," adding, "Physical AI should not be seen as a mere extension of LLMs, but rather as a separate issue altogether."

Accordingly, Doosan has chosen a "problem-narrowing strategy." Instead of building a universal humanoid capable of any task from the outset, the company aims to solve tasks that are difficult to automate and require significant human skill and training, one by one. Kim explained, "Rather than building a humanoid robot and then worrying about what to make it do, we first define task intelligence and then optimize the form factor."

The control methods for robots are also evolving. Traditional industrial robots required humans to define tasks and to handle teaching, setting coordinates, and programming the work process in detail. The "agentic robots" being developed by Doosan, however, only require a person to specify 'WHAT' to do; the robot then perceives the environment, creates a work plan, executes it, observes and verifies outcomes, and, if problems occur, autonomously replans.

Kim stated, "In the future, the primary operators of robots will be agents rather than humans," adding, "We are moving from an era dominated by robot programming to one where managing objectives becomes central."

This content was produced with the assistance of AI translation services.