Most industrial robots are still blind, deaf, and dumb. They follow a pre-programmed path, and if something is a millimeter out of place, they crash. The promise of vision AI has been to give them sight, but Sze Yuan Cheong thinks that’s only half the equation. His startup, Devol Robots, is betting the real unlock for automation isn’t just seeing the world, but feeling it [International Business Times (IBTimes UK), Feb 2025].
Founded in 2023, Devol is a small, quiet team split between San Francisco and Malaysia. They are not building another software layer to bolt onto off-the-shelf robot arms from Fanuc or ABB. They are designing the arms themselves, from the actuators up, with a physics-aware AI brain built in. The goal is a machine that understands contact, weight, and friction the way a human hand does, turning brute-force automation into something more like skilled labor [YouTube, 2024]. It’s a full-stack, deep-tech bet that tries to solve manipulation not with more data, but with better embodiment.
The bet on force, not just vision
The core of Devol’s pitch is that true robotic dexterity requires a sense of touch. Current systems use cameras and lidar to build a 3D map of the environment, then plan a collision-free path. This works for moving a box from point A to B, but it falls apart for tasks requiring pressure, like inserting a peg, polishing a surface, or assembling two parts with a tight fit. A vision-only robot will either miss the hole or smash the parts together. Devol’s approach is to embed force sensors directly into its proprietary actuators, creating a robotic hand that can sense resistance in real time and adjust its grip and motion on the fly [Teeming.ai, 2024].
The AI platform, which they call an "action layer for the physical world," uses this stream of force data to build a predictive model of physical interaction. The robot doesn’t just know where an object is; it learns how that object will behave when pushed, pulled, or squeezed. In a manufacturing context, this could mean a robot that can recover from a misaligned part without stopping the line, or one that can handle delicate, irregular items like food or textiles without crushing them [F6S, 2024]. The first deployments, according to the company, are in manufacturing cells, where this kind of adaptive, contact-rich manipulation is a bottleneck that pure speed can’t solve [YouTube, 2024].
Why a solo founder is building hardware
Founder Sze Yuan Cheong’s background is a mix of robotics engineering and, crucially, over a decade in manufacturing and industrial engineering [Grit Daily News, Unknown]. This isn’t an AI researcher trying to retrofit a theory onto a factory. It’s someone who has likely seen the limitations of current automation firsthand. The decision to build the hardware in-house, rather than license a software stack, is a massive undertaking for a pre-seed company. It speaks to a conviction that the sensing and the actuation must be co-designed; you can’t get true force awareness by strapping sensors onto a rigid, high-torque industrial arm built for repeatability, not sensitivity.
The team structure is lean and focused. Public records show Cheong as the solo founder and CEO, with recognition from other robotics founders suggesting he’s tapped into a technical network [LinkedIn, 2026]. The company is hiring for engineering roles, indicating a build phase centered on core R&D rather than sales [Jobstreet]. For a venture of this capital intensity, the path will depend heavily on proving a technical milestone that can attract the kind of funding needed to scale hardware production.
Where the wheels could come off
Building a new robotic arm from scratch is one of the hardest ventures in tech. The list of companies that have successfully gone from prototype to volume production in this space is short, and littered with expensive failures. Devol’s bet is a double one: first, that its force-control and AI software is meaningfully better than incumbents integrating third-party force-torque sensors, and second, that it can manufacture reliable, cost-effective hardware at scale. The competitive landscape is not standing still.
- The incumbent moat. Companies like Universal Robots have spent years refining collaborative arms (cobots) with built-in force sensing for safe human interaction. Their advantage is a massive installed base and a mature supply chain.
- The software wedge. A host of AI-focused startups are betting they can achieve dexterity through better vision models and simulation, avoiding the hardware grind entirely. Their argument is that cameras are getting cheaper and models are getting smarter, making touch a secondary concern.
- The unit economics cliff. Even with superior technology, Devol must hit a price and reliability point that makes a factory manager willing to swap out a known, dependable system for a newcomer. In manufacturing, downtime is measured in thousands of dollars per minute.
The counter-bet, then, is that vision and simulation alone will hit a fundamental wall with contact physics, and that retrofitting force sensing onto arms designed for stiffness will always be suboptimal. Devol is betting that the only way to win the manipulation game is to own the entire stack, from the metal fingers to the AI cortex.
Doing some back-of-the-envelope math illustrates the scale of the challenge. A high-end collaborative robot arm today can cost between $30,000 and $50,000. For Devol to compete, its proprietary hardware plus its "embodied AI" software likely needs to land in a similar band, while delivering a step-change in capability that justifies the switching cost. If their system can reduce line stoppages due to part misalignment by even a few percent, the payback period for a $50,000 arm in a high-throughput facility could be under a year. That’s the equation they need to solve: superior uptime must cover the premium of being an unproven brand.
Ultimately, Devol Robots isn’t just competing with other startups. Its real test is whether it can do something a Fanuc robot, programmed for one specific task for the last 40 years, fundamentally cannot. It has to build a hand that feels, and then convince the world that feeling is worth the price of admission.
Sources
- [International Business Times (IBTimes UK), Feb 2025] The Founder Behind Devol Robots, the Startup Rethinking How We Interact with Machines | https://www.ibtimes.co.uk/devol-robots-force-based-ai-automation-1781557
- [AiLaunchpad.my, 2024] Devol Robots | https://ailaunchpad.my/ai-solutions/devol-robots/
- [Teeming.ai, 2024] Devol Robots | https://teeming.ai/c/devol-robots/394882e9-2b95-f0ab-7e18-9ef0027ec100
- [F6S, 2024] Devol Robots Inc | https://www.f6s.com/company/devol-robots-inc
- [YouTube, 2024] The Future of Physical Intelligence | Devol Robots Interview | https://www.youtube.com/watch?v=i_ZBNUebh6I
- [Grit Daily News, Unknown] Devol Robots profile | https://gritdaily.com/devol-robots/
- [LinkedIn, 2026] Millie Yang - CEO and Cofounder at Breeze.com | https://www.linkedin.com/in/millieyang/
- [Jobstreet] Devol Robots job listings | https://my.jobstreet.com/devol-robotics-jobs