Why Do Humanoid Robots Still Struggle With the Small Stuff? | Quanta Magazine
To have robots which work like humans, I think we have to master physics.... Kim’s actuators got around the problem with controllable “compliance,” or flexible springiness. Over the past decade, they’ve gotten cheaper and more widely accessible.
Key takeaways
The most recent headlines show humanoid robots moving from laboratory prototypes to commercial deployments across several sectors. On March 18, 2026, Agibot demonstrated large‑scale reliability by staging a fully robot‑led live show, highlighting coordinated performance of dozens of units and confirming that more than 5,000 humanoids had been delivered worldwide by the end of 2025, signaling a shift toward repeatable, scalable supply. At the same time, IntBot announced that its general social‑intelligence engine now powers the Nilo concierge robot, which is operating 24 hours a day in three U.S. hotel chains and can handle multilingual guest interactions, underscoring a focus on hardware‑agnostic, socially aware services. In the consumer market, the San‑Francisco‑based startup Sunday secured a $165 million Series B round that values the company at $1.15 billion as it prepares to launch the household robot Memo for chores such as laundry and table clearing. Meanwhile, automakers continue to test humanoids in factories: BMW is trialling the Hexagon‑developed Aeon robot at its Leipzig plant, and Mercedes‑Benz is investing in Apptronik’s Apollo platform for parts‑moving and inspection tasks. Academic analysis published on March 13 notes that despite advances in vision‑language‑action models and compliant actuation, humanoids still struggle with fine‑motor manipulation of small objects, indicating that technical challenges remain even as commercial use expands.
To have robots which work like humans, I think we have to master physics.... Kim’s actuators got around the problem with controllable “compliance,” or flexible springiness. Over the past decade, they’ve gotten cheaper and more widely accessible. “Reinforcement learning solved a lot of the [bipedal] locomotion problem, but the hardware was the enabler,” Kim said. If reinforcement learning and compliant actuation were gifts to humanoid robotics, multimodal AI put a bow on it. Cut to now. Humanoids have apparently become so advanced that Tesla is mothballing some electric car models to make way for its Optimus humanoid robot, and start-ups are preselling android butlers with a straight face. Hype aside, I was genuinely curious: Did a paradigm shift happen in the field when I wasn’t looking? Surely today’s robotic bipedal marvels can ascend a few stairs and open a door without breaking a nonexistent sweat, something they famously struggled with a decade ago. I asked each researcher: Can your flagship robot — Boston Dynamics’ Atlas or Agility’s Digit, two of the most credible and pedigreed humanoids on Earth — handle any set of stairs or doorway? Roboticists once coordinated each movement with various hand-engineered algorithms, using equations to model the (simplified) physics of the robot. Now they train neural networks to act as “whole-body controllers” by running countless digital simulations of the humanoid. To have robots which work like humans, I think we have to master physics.... Kim’s actuators got around the problem with controllable “compliance,” or flexible springiness. Over the past decade, they’ve gotten cheaper and more widely accessible. “Reinforcement learning solved a lot of the [bipedal] locomotion problem, but the hardware was the enabler,” Kim said. If reinforcement learning and compliant actuation were gifts to humanoid robotics, multimodal AI put a bow on it.
Mentioned in this article