Lens Technology’s 37-DOF Hand Claim Is Technology News With a Manufacturing Test

Lens Technology has reportedly joined CASBOT to develop a dexterous robotic hand with 37 joint degrees of freedom. That specification makes this technology news worth watching, but it also creates an unusually demanding test.

A degree of freedom, or DOF, represents an independently movable direction in a mechanical system. More DOF can support richer finger poses, better object adaptation, and finer manipulation.

However, the headline number does not establish how many joints receive independent actuation. It also says nothing about tactile sensing, grip force, repeatability, heat, control latency, maintenance, or production yield.

The reported hand plan appeared on August 10, 2026. The underlying partnership developed over a longer period.

Lens led a financing round for CASBOT’s parent, Beijing Zhongke Huiling Robot Technology, in June 2025. The companies then became shareholders in a new embodied-intelligence venture formed in Hunan during 2026.

That sequence matters more than the specification alone. It connects a robotics developer with a manufacturer experienced in precision components, modules, and high-volume assembly.

The primary contest is therefore not Lens Technology against one named hand vendor. It is a contest between mechanical ambition and repeatable industrial production.

A 37 DOF robot hand can attract attention during a demonstration. A manufacturable hand must repeat useful work thousands of times while surviving impacts, contamination, cable wear, and imperfect objects.

The partnership becomes important if Lens can turn CASBOT’s research into dependable hardware. Until that happens, the 37-DOF figure remains a reported target rather than verified product performance.

What Lens Technology and CASBOT Actually Changed

The companies are combining robotics development with a manufacturing platform, not simply adding another hand prototype to the market.

Lens Technology is best known for precision components and assembly work across consumer electronics and other hardware categories. Its robotics strategy now covers parts, functional modules, and complete-machine manufacturing.

CASBOT develops general-purpose humanoid robots and embodied-intelligence systems. It grew from research connected with the Chinese Academy of Sciences and was established as a commercial company in 2023.

The relationship became financially concrete on June 26, 2025. CASBOT announced an angel-plus financing round of nearly RMB 100 million, with Lens Technology serving as the lead investor.

The funding was designated for production, product development, and market expansion, according to the financing disclosure. Existing and new investors also participated.

Lens later disclosed a RMB 50 million investment balance associated with Beijing Zhongke Huiling in its 2025 interim reporting. That filing provided a formal record of the relationship.

The next step arrived in 2026. Zhongke Huisi Embodied Intelligence, a Hunan company with registered capital of RMB 45 million, was formed with Lens and CASBOT’s parent among its shareholders.

Its registered activities include robot research, robot sales, artificial-intelligence hardware, and industrial robot manufacturing. The venture registration was reported on July 3.

The August 10 hand report therefore did not emerge from a new or informal collaboration. It followed investment, formal disclosure, and the creation of a jointly owned operating company.

What remains unavailable is equally important. No accessible official specification sheet has established the hand’s actuator count, sensor layout, weight, payload, or planned release date.

The phrase “37 joint degrees of freedom” also requires technical clarification. Joint freedom and actively driven freedom are not automatically the same measurement.

A hand can contain several movable joints while using fewer motors. Tendons, linkages, springs, or differential mechanisms can couple multiple joints to one actuator.

That approach can reduce weight, size, and component count. It can also limit the controller’s ability to command every joint independently.

The distinction changes the meaning of the headline. Thirty-seven independently actuated axes would create a different control and packaging problem from 37 articulated movements driven through coupling.

It is also unclear whether the reported number applies to one hand, a paired system, or a hand-and-wrist assembly. Those configurations should not be compared as equivalent products.

For now, the safest conclusion is narrow. Lens Technology and CASBOT reportedly plan a high-DOF hand, supported by an established investment and manufacturing relationship.

That is a meaningful change. The exact capabilities of the Lens Technology dexterous hand remain unverified until the partners release engineering documentation and reproducible demonstrations.

Why This Technology News Puts Robot Suppliers Under Pressure

The partnership raises the standard from designing an impressive hand to building one repeatedly, at acceptable yield, for identifiable customer tasks.

Dexterous hands sit at an awkward point in humanoid robotics. They need compact mechanisms, dense sensing, reliable wiring, and fast control inside a small, impact-prone package.

Each additional controlled axis can require another actuator, encoder, transmission path, bearing set, cable route, or control channel. Those additions increase integration difficulty.

The hand also sits at the robot’s point of contact with the world. Errors there can damage the robot, the object, or a nearby person.

A high-DOF design can provide more possible poses. It simultaneously expands the control space that software must manage.

That creates pressure for established hand developers pursuing simpler products. A lower-DOF hand must justify its design through cost, durability, force, or easier control.

It also pressures vendors selling research-oriented hands with high flexibility but limited manufacturing support. Buyers increasingly need repair procedures, replacement parts, calibration tools, and integration software.

Lens brings a different type of credibility to this competition. Its advantage is not a published manipulation benchmark for the reported hand.

Its advantage is experience industrializing compact hardware with demanding tolerances. That background can help with tooling, assembly processes, supplier qualification, and quality control.

Lens says its robotics operation already supplies head modules, joint modules, dexterous-hand modules, and precision structures to customers. Its 2025 annual report describes robotics as a core growth area.

The report also says head modules entered stable mass production. Other robot components entered domestic and international customer supply chains, according to the company.

Those statements support the manufacturing thesis, but they do not validate this particular hand. A supplier can mass-produce one module without solving every reliability problem in another.

CASBOT contributes robot-level integration and access to application environments. Its systems have targeted industrial, mining, inspection, research, and training settings.

That matters because hand development cannot be separated from the arm, controller, perception stack, and target task. A successful grasp depends on the complete system.

Consider a robot moving an irregular component from a bin. It must identify the object, select contact points, position the wrist, control force, and recover from a bad grasp.

Extra freedom helps only when software can use it. Poor calibration can turn additional joints into additional sources of error.

The same problem appears in delicate assembly. More articulation can help align connectors or manipulate flexible materials, but small position errors can still cause failure.

Manufacturers therefore face a forced choice. They can increase mechanical capability, or they can focus on a narrower set of tasks with simpler hardware.

The Lens-CASBOT proposal attempts to combine both routes. It aims for high articulation while attaching the project to a manufacturer positioned for scale.

That combination makes the announcement significant technology news. It challenges rivals to discuss production evidence alongside DOF, speed, or visually impressive hand gestures.

The pressure will become real only after Lens names the product, publishes specifications, and identifies a deployment path. Until then, rivals can reasonably treat it as an untested program.

The 37 DOF Robot Hand Has a Control Problem Before It Has a Sales Story

More joints expand the hand’s possible motion, but useful dexterity depends on actuation, sensing, software, and task success working together.

A human hand does not rely on joint count alone. Touch, proprioception, compliance, learned coordination, and rapid feedback turn anatomical movement into useful manipulation.

Robotic hands face the same systems problem with tighter packaging limits. Motors generate heat, transmissions introduce friction, and sensors consume space and power.

The reported 37 DOF robot hand will need a clear mechanical architecture. Common options include tendon drives, linkages, gears, direct drives, and hybrid systems.

Tendon mechanisms can relocate motors away from the fingers. This reduces hand mass and creates room for sensing.

However, tendons can stretch, wear, slip, and produce friction. Routing changes can complicate calibration and maintenance.

Linkage systems can create predictable motion and support passive load resistance. Their coupled movements can reduce independent control at specific joints.

Direct-drive architectures shorten the transmission chain. They demand small actuators with sufficient torque, acceptable heat, and compact power electronics.

No architecture removes every tradeoff. The commercial question is which compromise best supports the intended work.

Research published in 2026 illustrates the challenge. The open-source MM-Hand uses 21 DOF with remote tendon-driven actuation, multimodal sensing, and modular structures.

Its designers moved motors away from the hand to address space, mass, and heat. Their 21-DOF design demonstrates how one choice shifts complexity elsewhere.

Another 2026 project, DexLink Hand, uses 16 actuators to drive 20 joints through a linkage architecture. It emphasizes compactness and mechanically efficient motion.

The project reported successful reproduction of 33 established grasp types. That evidence is more informative than joint count because it connects design choices with task coverage.

Historical work offers another useful reference. A Chinese Academy of Sciences team developed several Casia Hand configurations for different markets.

One research-oriented model used 25 joint DOF and 21 driven DOF. A general-purpose version used 15 joint DOF with seven driven DOF.

A faster adaptive version retained 15 joint DOF but used between one and three driven DOF. The Casia Hand awards announcement described these distinct design goals in 2024.

That product family shows why two DOF figures should not be blended. Joint movement describes mechanical articulation, while driven freedom describes direct control authority.

The same clarification is essential for the Lens Technology dexterous hand. Otherwise, readers cannot compare it with systems using different counting conventions.

Sensing deserves equal scrutiny. A hand can close around an object using position commands, but reliable manipulation often needs force or tactile feedback.

Joint encoders report mechanism position. Force sensors estimate applied loads, while tactile arrays can detect pressure distribution, contact location, and slip.

Those signals must arrive quickly enough for the controller to respond. Slow feedback can crush a fragile object or lose a moving one.

Calibration also becomes harder as the mechanism grows. Small errors across many joints can accumulate at the fingertips.

Software then must map a desired task into coordinated joint commands. This process includes grasp planning, collision avoidance, force control, and failure recovery.

Data collection presents another constraint. Training a policy for one gripper is simpler than collecting broad demonstrations for a highly articulated hand.

More possible motion creates a larger action space. The robot needs enough examples to distinguish helpful finger motion from unnecessary complexity.

Simulation can reduce physical data requirements, but simulated friction and contact rarely match every real surface. Transferring policies to hardware remains difficult.

This is the central mechanism behind the story. The 37-DOF specification increases potential dexterity and simultaneously raises the cost of controlling it.

Lens can improve the manufacturing side through process engineering. CASBOT can integrate the hand with robot software and application data.

Neither contribution guarantees task performance. The companies must show that the complete system converts added motion into higher success rates or broader task coverage.

A benchmark should include unfamiliar objects, changing orientations, repeated cycles, and recovery from failed contact. A carefully staged demonstration cannot answer those questions.

What the 37-DOF Claim Does Not Yet Prove

The announcement establishes an ambitious direction, but it does not prove independent actuation, reliability, customer demand, or production readiness.

The first uncertainty concerns terminology. The public claim needs a diagram showing every counted joint and identifying which axes receive independent commands.

Without that diagram, comparisons with 16-DOF, 20-DOF, or 25-DOF products can mislead. Different vendors count wrists, palms, passive joints, and coupled movements differently.

The second uncertainty concerns physical integration. More actuators and sensors can increase mass, power use, heat, wiring density, and failure points.

A hand that performs well on a fixed test stand might behave differently at the end of a moving arm. Added distal mass affects arm dynamics and energy consumption.

The third uncertainty is durability. Industrial users care about cycle life, contamination resistance, impact tolerance, cable replacement, and recalibration time.

A robot working near machine tools can encounter oil, dust, sharp edges, vibration, and unexpected contact. Laboratory dexterity does not automatically survive that environment.

Serviceability can decide whether a product remains deployed. A modular finger that technicians replace quickly offers a different operating profile from a tightly integrated hand.

Lens has experience with manufacturing discipline, but the partners have not published yield or reliability data for this product. Those values cannot be inferred from corporate scale.

The fourth uncertainty is software maturity. A highly articulated hand requires stable low-level control and task-level policies that exploit its motion.

Software can also neutralize hardware advantages. If a controller uses only a small set of grasp patterns, many mechanical freedoms remain idle.

The fifth uncertainty concerns economics, even without discussing price. More components can increase assembly time, calibration effort, testing requirements, and support burdens.

Volume can reduce some manufacturing costs. It cannot eliminate complexity that appears during every calibration or repair.

The sixth uncertainty is customer fit. Many industrial tasks do not require a humanlike hand.

A two-finger gripper can outperform a complex hand when objects are uniform and the motion is repetitive. It can be easier to clean, validate, and replace.

Dexterous hands become more compelling when objects vary, workspaces change, or tools were designed for people. Examples include sorting mixed items and handling irregular components.

They can also help when factories want robots to use existing tools without redesigning every station. That benefit must outweigh the added operational burden.

CASBOT’s industrial relationships provide potential test environments. Still, an evaluation at a partner site is not the same as sustained customer adoption.

The article’s August 10 publication date also should not be confused with a confirmed launch date. The report describes an intended product, not a completed release.

No verified product name, technical manual, certification record, or customer delivery schedule was publicly established alongside the claim.

This verification gap does not make the story irrelevant. It changes the correct way to interpret it.

The announcement is evidence that Lens and CASBOT are targeting high-complexity manipulation hardware. It is not evidence that they lead the market in usable dexterity.

Competitors retain several response options. They can offer fewer freedoms with stronger force, lighter weight, better tactile sensing, or easier maintenance.

They can also specialize by application. A hand optimized for assembly can use different mechanics from one designed for household objects or hazardous inspection.

This is why the main contest remains promise against production evidence. The specification opens the contest, but it does not settle it.

Readers should also resist treating every high-DOF design as a substitute for every other design. A research hand, prosthetic hand, and factory hand face different constraints.

A serious evaluation needs an application, a test protocol, and a failure definition. Without those elements, DOF remains a design description rather than a business result.

Three Signals Will Decide Whether Lens Can Industrialize Dexterity

The next evidence should arrive in a strict order: engineering disclosure, repeatable task results, and customer-linked production data.

The first signal is a complete product release. Lens or CASBOT should publish a specification sheet that separates joint DOF from active DOF.

The sheet should identify actuators, transmissions, sensors, hand weight, payload, fingertip force, control interfaces, and supported operating modes.

A mechanical diagram would resolve whether the 37 freedoms belong to one hand, two hands, or a hand-and-wrist assembly.

This disclosure would strengthen the partnership’s case if the architecture supports substantial independent control without excessive size or mass.

It would weaken the claim if most movements are passive or tightly coupled while marketing presents them as independently controlled axes.

The second signal is a repeatable manipulation benchmark. The partners should demonstrate tasks that require the claimed articulation, not generic grasping.

Useful tests include object reorientation, connector insertion, tool use, mixed-item sorting, and manipulation under changing object positions.

The results should report trial counts, success rates, cycle times, object variation, and recovery behavior. Long demonstrations should disclose interruptions and maintenance.

An unfamiliar-object test would carry more weight than a memorized sequence. It would show whether the system generalizes beyond a curated demonstration.

This evidence would strengthen the 37 DOF robot hand thesis if additional freedom improves task coverage or success against simpler hardware.

It would weaken the thesis if the hand uses only a few fixed grasp patterns or fails when friction, position, or object shape changes.

The third signal is customer-linked production evidence. Lens should identify whether the hand enters pilot manufacturing, internal deployment, or external delivery.

Relevant evidence includes shipment categories, qualified production lines, yield trends, field hours, repair rates, and repeat orders.

A named customer is not essential if confidentiality prevents disclosure. However, Lens can still describe the application, deployment stage, and acceptance criteria.

The company has said its broader robotics business already supplies modules and complete-machine services. The question is whether this specific hand joins that production flow.

Pilot units operating inside Lens factories would offer a logical first proving ground. Such use would connect the partnership’s robotics software with Lens manufacturing conditions.

Internal deployment should not count as automatic validation. The company would still need to explain task performance and operational savings.

External repeat orders would provide stronger evidence because another organization would be accepting the maintenance and integration burden.

These three signals also tell competitors how to respond. A credible specification invites technical comparison, while a credible benchmark shifts attention toward software and sensing.

Customer-linked production would raise the stakes again. It would show that the hand survived procurement, integration, and daily operation.

The absence of those signals would leave the story at the announcement stage. Investors and buyers should then avoid treating 37 DOF as an established commercial advantage.

For developers, this technology news highlights a broader shift in embodied AI. Hardware architecture, training data, control software, and manufacturing now need joint evaluation.

For enterprise buyers, the lesson is practical. Ask which tasks improve, how results were measured, and what happens after a finger or sensor fails.

For other hand vendors, the Lens Technology dexterous hand creates pressure to publish comparable evidence. A lower joint count can remain competitive when it delivers better uptime or task economics.

The partnership has a plausible division of labor. CASBOT brings robot development and application integration, while Lens brings precision manufacturing and supply-chain execution.

Its unresolved challenge is equally clear. The companies must convert a large motion space into dependable manipulation without letting complexity overwhelm the product.

That is the real reason this story matters. It is not simply another contest to publish the largest number beside a robotic hand.

The next useful update should answer engineering questions that the headline leaves open. Watch for the active actuator count, repeated task data, and pilot deployments.

Until those arrive, treat 37 DOF as a design target backed by a serious manufacturing partnership. Do not treat it as proof of superior dexterity.

The decisive question is straightforward: can Lens and CASBOT show a hand that performs difficult work repeatedly, then build and service it at scale?