By Sarah Xiao 
A patent application from a Fudan University team, published this week under the title “A Bio-Inspired Underactuated Dexterous Hand,” attacks a trade-off that has shaped the category for a decade: build a cheap hand with few motors and it can only grip; build one that can actually manipulate and you end up with a motor per joint, a controller per motor, and a bill to match.
The Fudan design does not put a transmission in each finger. It puts a three-degree-of-freedom parallel platform on the back of the hand, and lets that one platform schedule the flexion and abduction of all four fingers. Five servos in total. Fifteen degrees of freedom. The fingers cannot move independently — and that is the point, not the compromise.
The cost of one tendon per finger
A conventional underactuated hand gives every finger its own tendon and its own actuator. It is easy to build and easy to reason about. It also means the fingers have no mechanical relationship with each other: nothing makes them curl in sequence, nothing makes them close slowly around an object the way a human hand does. Most designs go further and delete lateral movement entirely, because a finger that can swing sideways needs a second transmission. The result grips flat, regular objects and gives up on sheets, thin parts and anything with an awkward shape.
The Fudan filing keeps both motions and removes the per-finger drives. It routes all four fingers into one triangular platform on the back of the hand, and splits the transmission into two routes that do not touch each other: cable tubes handle flexion, spring plates handle abduction.
Three motions, one platform
Because the platform has three degrees of freedom, it can do three useful things to the hand, and each one maps to a grasp primitive.
Translate along the fingers. All four fingers flex together. This is the power grasp: a stable, even wrap around a regular object.
Translate across the fingers. All four fingers swing sideways together. This is the pinch-and-sheet grasp — thin parts, odd shapes, the cases a pure flexion hand drops.
Rotate. This is where the design gets interesting, and it is worth slowing down on.
Rotation does the sequencing
The four tendons do not terminate at the same radius from the platform’s rotation centre. When the platform turns, each tendon is pulled by a different amount — the index finger takes up more cable than the middle finger, the middle more than the ring, the ring more than the little finger. The curl therefore arrives in order, from the index outward, without a single line of sequencing code.
The abduction route does the same thing in a different way: the spring plates are offset from each other, so as the platform rotates they displace progressively and the fingers splay a little more with each step.
The consequence is not cosmetic. Sequencing a grasp in software means a state machine, position feedback and timing for every finger; sequencing it in geometry means the mechanism produces the order by construction. For an embedded controller on a hand that has to hit a price point, that is the difference between a control problem and a lookup table.
The thumb: a cable through the centre of a sphere
The thumb is where underactuated hands usually fall apart. When a tendon-driven thumb rotates into opposition, the cable path lengthens and shortens as the thumb turns, so the flexion posture drifts the moment the thumb starts to rotate. Pinch becomes unreliable exactly when it is needed.
The Fudan filing uses a spherical crossed four-bar as the thumb’s rotation mechanism, and routes the flexion tendon through the centre of that sphere. Geometrically, a cable passing through the centre of rotation does not change length when the sphere rotates — so the thumb can swing through a large opposition range (the filing quotes up to 235 degrees) without disturbing the flexion it has already set. Two servos run it: one for opposition, one for flexion. The old interference between the two motions is designed out rather than compensated for.
The finger itself: two crossed four-bars and a flexure hinge
Each finger is two sets of crossed four-bar linkages driven by a single tendon. Pulling that tendon curls the fingertip first, then the middle phalanx, then the base — the order a human finger closes in, contact first and wrap after.
The structure is printed in one piece from two materials: PLA for the rigid links, PEBA for the hinges. The flexure replaces both the return spring and the pin joints, which removes the clearance and stiction that pin joints bring and the parts count that springs add. The trade-off is on the process side: two materials fused in one print are sensitive to print parameters and filament quality, and the fatigue life of a PEBA hinge under repeated bending is exactly the thing a paper cannot settle. It needs a prototype, a cycle tester and time.
What five servos buy
Three servos drive the platform, which handles every coordinated action of the four fingers. Two servos drive the thumb. Five actuators, fifteen bio-inspired degrees of freedom — three degrees of freedom per actuator.
That ratio is the whole argument. A fully actuated dexterous hand spends one motor, one driver channel and one control loop per degree of freedom. The Fudan hand spends one actuator per three, and the control law collapses to the inverse kinematics of a planar parallel mechanism: a closed-form geometry problem, not a multi-finger planning problem.
The rest of the design points the same direction. The palm and fingers print on a standard FDM machine with dual extrusion, so a prototype is a print rather than a procurement cycle. The wrist mounts to a standard end-effector flange, so the hand can go onto an existing arm without an adapter project.
Where the design gives something up
The four fingers are mechanically coupled, and that is not a side effect — it is the mechanism. They cannot be commanded independently.
Anything that needs a single finger to work alone is out of reach: poking a button with the index finger, typing, a one-finger fine adjustment while the other three stay put. A hand with independent fingers is more flexible in that sense, and any application built around single-digit precision should not choose this one.
The honest reading is that this is a hand for a specific job: cheap, able to roll an object inside the palm, able to adapt to irregular shapes, and not required to do the finest single-digit work. That is a large share of service robots and mobile manipulators, which is the segment the design seems aimed at.
Why it matters beyond one lab
The dexterous hand market has been split down the middle. At the top, fully actuated hands with twenty-plus degrees of freedom and a price to match, sold into research and a handful of high-value industrial cells. At the bottom, two-finger grippers and simple underactuated hands that grip and release and nothing more.
The gap in between — a hand that can re-orient an object in the palm, adapt to a shape it has never seen, and cost a fraction of a fully actuated unit — is where service robotics, logistics and mobile manipulation actually live. Most attempts to fill it add fingers or add sensors. This one subtracts actuators and puts the intelligence in the linkage.
Whether it works is now a manufacturing and durability question rather than a kinematic one. The kinematics are on paper and they are clean.
Three things to watch
One: whether the PEBA hinge survives. A flexure hinge that replaces a pin joint is only better if it lasts. The number to look for is cycles to failure under load, from a real prototype, not a simulation.
Two: whether print repeatability holds. Two-material FDM is sensitive to printer, filament batch and orientation. A design that only works on the team’s machine is a paper design. The question is whether a third party can print the same hand and get the same stiffness.
Three: whether the coupled grasp is enough for the buyers. If service-robot integrators need one finger to do something alone, the architecture is disqualified regardless of cost. If they mostly need a hand that wraps, rolls and adapts, five servos is a very short bill of materials.
The patent is a filing, not a product, and filings are where mechanisms go to be judged. This one is worth judging on a specific claim: that grasp sequencing is a geometry problem, and that a platform on the back of the hand can solve it more cheaply than five transmissions inside the fingers.