The Real Gap in Robotics is Innovation in Foundational Tools, Says NUS Professor Shi Fan

Robot Researcher Calls for a New Generation of Tools, Not Just Smarter AI
Subheading: NUS Professor Shi Fan argues that true robotics breakthroughs are stalled by a lack of innovation in foundational tools and urges the development of rigorous safety standards.

SINGAPORE – While the robotics industry is experiencing unprecedented growth and excitement, Professor Shi Fan of the National University of Singapore (NUS) warns that a critical bottleneck threatens long-term progress: a severe lack of innovation in the field's fundamental tools.

“If everyone shares the same set of underlying tools—the same simulators, platforms, and frameworks—then achieving truly disruptive innovation is almost impossible,” Professor Shi explained. “Everyone is holding the same ‘hammer,’ so it’s very difficult to knock out something completely different.”

This perspective is shaped by a global academic career that mirrors the evolution of modern robotics. From his start at Peking University during the early days of self-driving cars, to research stints in Tokyo and Zurich, and now leading the Human-Centered Robotic Lab (HcRL) at NUS, Shi has a unique vantage point.

He believes academia's role is to pursue research that is both groundbreaking and practical. "If a direction is not bold enough, it’s better left to industry," he states, adding that novelty for its own sake is a dead end without long-term value.

Reflecting on the past decade's changes, Shi highlights the "democratization" of robotics, driven by affordable hardware from companies like Unitree and DJI, and open-source software like the Robot Operating System (ROS). This has dramatically lowered barriers to entry and accelerated development.

As industry rapidly adopts complex AI techniques like reinforcement learning (RL), Shi sees a new challenge: ensuring the safety of these "black box" systems. His lab's past work on deliberately causing robots to fail through adversarial attacks was aimed at proactively identifying risks, similar to crash-testing in the auto industry.

He believes mandatory safety standards for robots, especially those in homes, are inevitable and may be more difficult to achieve than improving intelligence itself. "A robot must not cause harm to a human under any circumstances," he asserts.

On the topic of humanoid and general-purpose robots, Shi is cautiously optimistic. He sees strong value in a human-like upper body with two arms for complex tasks but believes the need for legs is more dependent on the specific application. True "generality," he suggests, is likely a robot being a "multi-talented specialist within a limited task space," like the impressive Dyna Robotics demo that could fold a wide array of random clothing.

Ultimately, Professor Shi concludes that the next great leap in robotics depends on innovating the underlying tools themselves. It is in this foundational, less-glamorous work that his team is focused, forging the new "hammers" needed for the future.