A sub-$150 three-finger hand that sells on 'good enough'

Let me say something. When I first saw the "thousand-yuan class" label on the DexHand021 S, I didn't quite believe it. In the dexterous hand business, a few thousand yuan won't even buy you a decent end effector, let alone one with actuation. Later I went through the Lingqiao Zhineng launch materials and understood—they first figured out which 80% of hand functions are actually useful in general scenarios, and just dropped the remaining 20%.

I buy this approach. Three fingers, 8 degrees of freedom, 22N grip force, with rigid-flexible coupled transmission. The specs are right there, and compared to their own DexHand021—the five-finger flagship with 19 DOF and 23 sensors—it's not even close.

But the thing is, most grasping tasks don't need 19 DOF at all. Grabbing a part on a production line, picking a box on a sorting line, demonstrating opening a door in an education setting—three fingers are enough. You pay for the extra DOF, and then you have to write an extra layer of control strategy for them.

And their attitude toward dexterity. At the launch they proposed a quantifiable dexterity evaluation metric and pointed out the pain points of current VLA frameworks. Mainstream vision-language-action models basically reduce action output to grasp pose generation, with very limited support for fine manipulation like in-hand operations—in-hand demonstration data makes up less than 1% of open-source datasets. That's a pretty honest point. A low-DOF hand with clear task boundaries can actually land sooner than a high-DOF hand with fuzzy control.

My judgment in one sentence: the value of this hand is that it pulls dexterity back from a show-off metric to task matching. At the thousand-yuan price point, what you're buying is sufficient certainty.

Still, I have one question I haven't resolved: where exactly is the ceiling for three fingers on in-hand manipulation—like turning a screw or flipping something over? Is it genuinely incapable, or has nobody written a strategy for it yet? I'm still testing this; I'll report back when I have conclusions.