Robotics has progressed rapidly in the past few years, but major obstacles — including data collection and trust infrastructure — remain barriers to widespread deployment. This was the takeaway from a recent panel of robotics experts at SuperAI Singapore, where they discussed the present and future of the industry.
The panel, moderated by Omdia chief analyst Lian Jye Su, noted that while humanoid robots attract much media attention, the most immediate practical applications lie in non-humanoid robotics: facilities management, logistics, and care provision in aging societies facing labor shortages.
"Personally, I think within three years, the average commercial building will have at least 6 to 10 robots," said Alan Ng, founder and CEO of QuikBot Technologies. Ng noted, however, that robot capabilities remain far short of replacing people, with high costs representing another barrier. That said, the field has made impressive improvements in recent years, as robots are becoming smaller, more robust, and able to run for longer periods of time.
According to Yanliang Zhang, chief scientist of Weston Robot, the greatest challenge 15 years ago was making sure robots did not fall down and could operate for longer than 45 minutes. "Today, most of our robots can run for around 6 hours. And we can put a solar panel on it, we can put a wireless charger on the foot. While it's walking or sitting, it can be charged," he said.
These improvements are at least partly attributed to breakthroughs in hardware — such as battery life and motor controls. Meanwhile, the AI layer — including foundation models, large language models (LLMs), and world models that allow robots to navigate complex environments — has also progressed significantly in the past few years.
The "golden triangle" of robotics
The panelists discussed the need to build a trust infrastructure before physical AI deployments, such as robots, can be scaled. Legal frameworks need to be established, such as determining liability in the event that a robot causes harm when deployed. Safety standards also need to be developed for physical AI to earn trust — not just from governments and the public, but also from enterprises seeking to roll out robots safely.
A "golden triangle" within robotics combines three core elements for scaling robotics in the real world: cheaper hardware, powerful AI models, and data. The last of these, however, is not always easily accessible due to data privacy or national security regulations.
This may lead to a resurgence in federated learning, according to Sony Research's senior executive director Michael Spranger. This is when multiple robots learn a skill without sharing data — instead processing it separately and sharing only their insights.
Ng predicted that the domestic robot industry will begin to take off within the next three to five years to perform undesirable household tasks such as cleaning. Spranger, on the other hand, predicted that robots will begin to have "character", especially when successfully combined with a proprietary personality or persona.
These predictions come as aging societies have made the development of physical AI more pressing than ever. The panelists predicted the sector to be worth roughly US$25–30 billion by 2050, as older populations mean fewer workers available for the manufacturing, healthcare, and elder care sectors.
Article edited by Jerry Chen
