At RoboBusiness this year, one of the panelists referenced Moravec’s Paradox — the idea that robots are great at tasks humans find hard like precision at scale. But, they struggle with tasks a toddler can do, like picking up a grape.
The Wall Street Journal just ran a piece titled “The ‘Hands Problem’ Holding Back the Humanoid Revolution.” It’s a phrase I’ve heard my son — a roboticist — use more than once, and when it shows up both in a lab and on the front page, it’s probably worth unpacking.
Because amid all the hype about walking, talking humanoids, there’s a quieter truth:
Robot hands are still a bottleneck.
What Is the Hands Problem?
Robots can walk. Robots can dance — for reasons unclear. They’re already lifting crates in factories and hauling parts. But ask one to help someone get dressed, make a sandwich, or change a lightbulb? These are routine, everyday tasks. And they’re still beyond what most machines can reliably do.
As the WSJ put it:
“Before they’re ready to turn a wrench, [robots] must solve … what Elon Musk calls ‘the hands problem.’”
Kevin Lynch, director of Northwestern’s Center for Robotics and Biosystems, offered this timeline:
“We’re setting 10 years as our goal to have dexterity, be functional and useful and able to do some of the things that humans do.”
Others in the field are more optimistic — particularly those exploring hybrid control systems that combine machine learning with human teleoperation. Companies like Sanctuary AI are building humanoids specifically designed for this model, while teams at Covariant — though focused primarily on warehouse picking — reflect a similar philosophy in how they blend perception, planning, and adaptation. Intrinsic, Alphabet’s robotics software arm, has also emphasized this approach in both its public demos and research, aiming to make robotic manipulation more robust by learning from human-guided examples.
🧠 Why It’s So Hard
Hands are deceptively complex. They involve:
20+ degrees of freedom in the fingers and wrist
Tactile sensing to detect slippage, pressure, texture
Adaptive control to handle objects of different weights, fragilities, and orientations
Fast reflexes to correct in-motion mistakes
“A robotic hand must make trade-offs between strength, dexterity, slenderness and ruggedness. Increasing one attribute can diminish another.”
— Alberto Rodriguez, Boston Dynamics (WSJ)
Even advanced robotics systems still struggle with this blend of strength, softness, and feedback — especially outside controlled environments.
🎯 Precision at Scale vs. Dexterity
Robots are masters of precision at scale. On assembly lines, they can place microchips with sub-millimeter accuracy — flawlessly, thousands of times per hour.
But ask that same robot to pick up a grape without squishing it?
That’s incredibly difficult for a robot. It’s the paradox: robots excel at things humans find hard, and struggle with the things a toddler can do.
🛠️ Who’s Working on It?
The hands problem remains a major focus — and multiple companies are racing to solve it, each in their own way:
🦾 Figure AI unveiled Figure 03 (announced late 2025), featuring tactile fingertips, palm cameras, and a compliant grip system designed for adaptive manipulation. Earlier prototypes like Figure 02 laid the groundwork, but 03 marks their clearest step toward real-world dexterity.
🔋 Tesla’s Optimus has demonstrated basic manipulation — folding laundry, opening a bottle — though these actions remain pre-programmed and brittle under variation. Its current hands appear optimized for strength and task-specific automation over reflexive dexterity.
🧠 Sanctuary AI’s Phoenix includes five-fingered hands and operates via a hybrid control stack blending human teleoperation and automation, emphasizing labor-capable dexterity.
🔄 Apptronik’s Apollo takes a modular approach, swapping out hands depending on the task — from simple hooks to articulated digits.
🕷️ Shadow Robot Company remains a go-to in research robotics, offering 20-DOF hands with advanced tactile sensing used in manipulation studies.
🧪 UBTech’s humanoids have been seen performing delicately staged tasks blending dexterity with theatrics.
🧬 Clone Robotics experiments with tendon-driven artificial muscles that mimic the fluid motion of biological hands, though still early-stage. (Although, Clone Robotics seems to always win in the category of “creepiest robot.”)
1X Technologies has demonstrated meaningful progress in hand and arm manipulation with its NEO Gamma robot, including object picking in varied environments and tendon-driven actuation, though fully general-purpose dexterity and detailed performance metrics remain ongoing challenges.
Everyone’s approaching the problem differently — through hardware, sensing, learning, materials — but the goal is the same: a robotic hand that can grasp, feel, adjust, and act in the real world.
🚪 Why It Matters
The hands problem isn’t a niche technical hurdle — it’s the threshold of usefulness.
Without dexterous manipulation, general-purpose humanoids remain showpieces. Solving it would unlock capabilities in:
Elder care, with gentle, safe interactions
Factories, with flexible tool use
Homes, with chores, cleanup, and cooking
Disaster response, where humans can’t safely go
Until then, robots might walk beside us — but they won’t be ready to work with us.
Final Thought
Forget the backflips. The true test will be whether a robot can turn a key, tie a knot, or hand you a mug without breaking it.
Vocabulary Key
Dexterous manipulation – Coordinated finger motion to adjust, rotate, or use an object with precision.
Degrees of Freedom (DoF) – The number of independent movements a joint or system can make.
Tactile sensing – Detecting touch, force, or texture via sensors.
Soft robotics – Robots made of flexible, compliant materials, often bio-inspired.
Moravec’s Paradox – A principle in AI: tasks that are hard for humans are often easy for machines, and vice versa.
FAQs
Why not just use simple grippers?
Grippers work well in structured environments, but fall short in everyday, dynamic tasks.
Which companies are closest?
Figure AI, Shadow Robot, and others are making progress, but no one’s cracked general-purpose, robust manipulation yet.
Why does it matter?
Because without useful hands, humanoid robots can’t help where they’re most needed: in homes, hospitals, or workplaces.
#RoboticHands #HumanoidRobots #DexterousManipulation #TheHandsProblem #RobotDexterity #thedroidsnewsletter #dianawolftorres #robotics
