LG CNS gathers 100+ robotics experts to push physical AI into factories

LG CNS packed more than 100 researchers and industry professionals into its Future Robotics Lab headquarters in Magok, Seoul, for a forum dedicated to one question: how does physical AI escape the laboratory and start working in factories, warehouses and logistics yards.

The event, the company's 2026 Robotics & Physical AI Forum, ran on a straightforward premise. The building blocks of physical artificial intelligence — robot foundation models, training data, intelligent robotic hardware and simulation — are advancing quickly in academia. Deployment on real industrial floors is not keeping pace. LG CNS wanted both camps in one room to close that gap.

The guest list reflected that ambition. Attendees included faculty members, graduate students and company employees from the University of Toronto, Sogang University, Chung-Ang University and Korea University. The mix matters. Foundation model research lives in university labs; the integration headaches live on factory floors. LG CNS sits in between, and it convened the forum to connect the two.

Why LG CNS thinks it can host this conversation

The Korean IT services giant did not arrive at the forum as a neutral convener. LG CNS said it has more than two decades of experience in industrial automation and has completed over 1,000 smart factory and smart logistics projects. That track record is its credibility card in the physical AI race, and the company played it openly during the event.

The stakes are concrete. Robot foundation models — large learned systems that can generalize across tasks, much as language models generalize across text — promise robots that adapt rather than repeat pre-programmed motions. Turning that promise into uptime on a production line requires solving problems that no single research group owns alone. LG CNS presented its own approach to bringing robotics technologies into industrial environments, grounded in those 1,000-plus delivered projects.

What are PhysicalWorks Forge and Baton?

The forum doubled as a showcase for two platforms LG CNS launched this year.

• PhysicalWorks Forge — a robot-learning platform, the layer where robots acquire new skills and behaviors.
• PhysicalWorks Baton — an agentic AI-based platform for robot operations and management, handling the day-to-day orchestration of robot fleets.

The pairing sketches the company's thesis. Learning gets robots capable; agentic management keeps them running. Both pieces have to work before a warehouse can treat robots as infrastructure rather than pilot projects.

What did the University of Toronto present?

The forum's research centerpiece came from Professor Lee Chi-geun of the University of Toronto, who presented on the development of robot foundation models and the technologies needed to deploy them in physical robots.

His presentation covered practical techniques for making learned robot motion usable. Lee discussed methods for combining data from different types of robots — a critical problem, because models trained on one machine's data often fail to transfer cleanly to another. He also addressed flow matching and action chunking, two techniques designed to help robots generate smoother, more continuous movements instead of the jerky, discrete steps that betray a naive learned controller.

The presentation was candid about limits. According to Lee's talk, differences in robot movement, latency and data availability remain among the central challenges facing the technology. A model that performs well on one robot arm can struggle on another with different dynamics, slower response times or thinner training data. Those gaps between simulation and steel are exactly where physical AI projects stall.

How did academia and industry align?

Beyond the formal presentations, participants exchanged views on several open fronts.

• Applications in manufacturing and logistics, the two domains where LG CNS has built its 1,000-project record.
• Operational challenges that emerge only after robots leave controlled test environments.
• The capabilities that companies and universities each need to develop to make physical AI deployable at scale.
• How human roles may evolve as robotics advances across industrial settings.

That last item is not a throwaway. As robots take on more physical work, the division of labor between people and machines shifts, and both corporate training programs and university curricula have to shift with it. The forum put that question on the same agenda as model architectures — a signal that LG CNS views workforce evolution as part of the deployment problem, not an afterthought.

What does this mean for the market?

Physical AI has become one of the most competitive frontiers in the technology industry, with major players across robotics, cloud and AI racing to pair foundation models with machines that act in the physical world. For an IT services firm like LG CNS, the opportunity is not necessarily in building the model but in the integration layer — taking research-grade techniques and hardening them for factory and logistics customers who measure success in throughput and downtime.

The company's positioning leans on that intermediary role. Two decades of smart factory work give it customer access and domain knowledge. University partnerships, of the kind on display in Magok, give it a pipeline to frontier research. The forum was designed to fuse the two.

Lee Jun-ho, executive vice president and head of LG CNS' Smart Logistics & City Business Division, framed the goal in a statement: "We hope to strengthen our technological competitiveness through continued engagement with academia and build our business capabilities so research results can lead to innovation in actual industrial settings."

The quote is a roadmap. Continued engagement with academia, then business capability building, then industrial innovation — in that order.

What comes next?

LG CNS said it hopes the forum format continues, with ongoing academic engagement turning research output into deployable industrial technology. The two PhysicalWorks platforms launched this year give the effort a concrete product surface. The open question, which the forum itself acknowledged, is whether robot foundation models can bridge the gaps in movement, latency and data availability fast enough for the 1,001st smart factory project to look fundamentally different from the first thousand.