By RWT Robotics Editorial Team Checked against 8 sources Published
The parts for Hugging Face’s open-source SO-101 robotic arm cost about $130. A Universal Robots UR7e cobot lists at up to $48,682 from a US distributor. Price follows stiffness, precision, payload, software and support far more than axis count, so this guide sets out the seven price tiers with US listed prices and where each model is sold, then covers the specs and hidden costs that decide whether an arm can do your job.
Key takeaways
- Robotic arms sell in seven price tiers, from under $100 toy kits to premium cobots at $38,000โ$49,000.
- The biggest jump sits between budget industrial cobots ($6,999โ$9,500) and premium cobots, and it pays for ecosystem, service and safety documentation.
- The listed price is never the system price. Grippers, cameras, mounting, safety and integration can double or triple a cobot project.
- Under ISO 10218:2025, “collaborative” describes an application that needs a risk assessment. It is not a safety label on the robot.
- An open Python SDK and maintained ROS 2 support keep an arm useful for longer than one extra axis does.
Fact-checked
We checked these key facts against the sources named, on November 16, 2026:
- Companies ordered 36,766 robots worth $2.25 billion in 2025, and cobots made up 19.6% of the units (source: businesswire.com)
- The US passed Japan for second place with about 38,400 units, up 12%, while Japan fell 19% to 36,219 (source: manufacturingdigital.com)
- ISO 10218 is the international safety standard for industrial robots, and the 2025 revision is the biggest update since 2011 (source: therobotreport.com)
- According to A3, cobots made up 28.6% of all robots ordered in North America in Q4 2025 but only 10.7% of full-year order revenue (source: automate.org)
Robotic arm prices at a glance: the 7 tiers
A robotic arm costs under $100 as a toy kit, $649โ$1,014 as an assembled desktop educational arm, $6,999โ$9,500 as a budget 5 kg industrial cobot, and $38,000โ$49,000 as a premium cobot. Pick the tier by what the arm has to do, then pick the model with the best software in that tier.
Prices checked: November 2026. All figures are listed USD prices before sales tax, taken from maker stores and authorised US resellers. Shipping, tariffs and bundles change them.
| Tier | Example models | Listed price (USD) | Typical payload | Precision reality | Software | Where to buy (US) | Best for |
|---|---|---|---|---|---|---|---|
| 1. Toy and STEM kits | Teach Tech Hydrobot, hydraulic and lever kits | Under $100 | Grams | None to speak of | None, or manual | Amazon, toy and STEM retailers | Kids, first mechanics lesson |
| 2. Hobby servo arms | Hiwonder kits, Waveshare RoArm | $100โ$400 | ~100โ500 g near the base | Millimetres of play, jitter | Arduino, ESP32, Python | Hiwonder, Waveshare, Amazon | Learning electronics and kinematics |
| 3. Open-source AI arms | Hugging Face LeRobot SO-101 | $130โ$240 | Light objects | Good enough for imitation learning | LeRobot, Python | Seeed Studio | Collecting AI training data |
| 4. Desktop educational arms | mechArm 270, myCobot 280, DOBOT Magician Lite | $649โ$1,014 | ~250 g | Around ยฑ0.5 mm class | Blockly, Python, ROS/ROS 2 | Elephant Robotics store, Stemfinity, School Specialty | Classrooms, ROS 2 learning |
| 5. Research and desktop industrial arms | UFactory Lite 6, DOBOT MG400, DOBOT M1 Pro, Niryo Ned2 | $2,999โ$8,995 | 300 gโ1.5 kg | Real encoders, better stiffness | Python, ROS 2, MoveIt | UFactory store, Trossen Robotics, RobotLAB | Labs, universities, light benchtop work |
| 6. Budget industrial cobots | Fairino FR5, DOBOT Nova 2, UFactory xArm 6 | $6,999โ$9,500 | 2โ5 kg | ยฑ0.02 mm claimed (FR5) | Vendor SDK, ROS 2 | Top3dshop, Dobot US store, Digi-Key, UFactory store | Small-business machine tending |
| 7. Premium cobots | Universal Robots UR7e, Franka Research 3 | $38,000โ$49,000 (UR7e); FR3 quote only | 3โ7.5 kg | ยฑ0.03 mm class | Huge ecosystem, documented safety | Vention, Automation Distribution, integrators | Production lines, research labs |
Which tier fits, in a line each:
- Parent or teacher of young kids: a tier 1 hydraulic kit. It teaches levers and pressure, not programming.
- Adult learning Arduino: a tier 2 servo arm under $300, as long as you accept the precision limits.
- AI or embodied-learning researcher on a budget: a tier 3 SO-101 leader and follower pair, plus a camera.
- University course or ROS 2 learner: a tier 4 desktop 6-axis arm with published URDF files and MoveIt configs.
- Lab or electronics bench: a tier 5 arm such as the Lite 6 or a DOBOT SCARA.
- Small factory: a tier 6 cobot rated for 5 kg, together with a proper risk assessment.
How much does a robotic arm cost in 2026? Tier by tier
Each tier up buys stiffness, precision, payload, software maturity and support. The tier table above has the headline bands. This section shows what each tier actually gets you, which models sit in it and where US buyers order them.
Tier 1: Toy and STEM kits (under $100)
These are hydraulic, lever-driven or simple motorised kits, such as the Teach Tech Hydrobot, which moves its joints with water-filled syringes. They are building projects. They teach how pressure and linkages turn into motion, and they do that well.
What you give up: almost everything a robot does. There’s no programming, no feedback and no repeatability. Buy one for a 10-year-old’s curiosity, not for a robotics course.
Tier 2: Hobby servo arms and kits for adults ($100โ$400)
This tier covers acrylic or aluminium-bracket arms with PWM hobby servos, usually driven from an Arduino, ESP32 or Raspberry Pi. Hiwonder and Waveshare (the RoArm line) are familiar names, along with countless unbranded kits on Amazon. For learning wiring, servo control and the maths of inverse kinematics, nothing gives you more for the money.
Reviews and forums keep reporting the same weakness: “too slow, weak and required you add all your own sensors.” That describes the tier accurately, and the causes are mechanical:
- Potentiometer feedback has a deadband of a few degrees, so the servo can’t hold an exact angle.
- Acrylic frames flex under load. A millimetre of bend at the shoulder becomes several millimetres at the gripper.
- Gear backlash in cheap metal-gear servos adds play at every joint, and the play stacks up along the chain.
- No joint state reaches the computer. Most PWM servos never report their position, so your code doesn’t know where the arm really is.
If you’re choosing between the Tau arm (~$250) and the Waveshare RoArm (~$190) for learning, check one thing first. Does the arm report joint positions and expose an API you can script, through a serial or Wi-Fi JSON interface or a Python library? An arm that reports its state teaches you closed-loop control. One that only accepts PWM commands teaches you wiring.
Tier 3: Open-source AI arms, the LeRobot SO-101 ($130โ$240)
The SO-101 (also listed as SO-ARM101) is an open-source, 3D-printable arm designed for Hugging Face’s LeRobot library. Hugging Face estimated the non-printed parts of one arm at about $130. At its 2025 launch, Seeed Studio sold a leader-plus-follower kit for $220 (Standard) or $240 (Pro). The design descends from earlier low-cost arms such as the ROBOTIS KOCH v1.1, which used Dynamixel servos in the same leader/follower layout.
The price overlaps with tier 2, but the hardware is a different class. The SO-101’s Feetech smart bus servos report their position digitally, which a PWM kit can’t do.
Why two arms? LeRobot’s standard workflow is teleoperation for imitation learning. You move the passive “leader” arm by hand, and the motorised “follower” copies it in real time while cameras record the scene and the joint angles. Those recordings become training data for a policy, a neural network that later drives the follower on its own.
Quick steps: an SO-101 imitation-learning setup
- Buy or print parts for two arms: a leader and a follower.
- Assemble them and set each smart servo’s ID.
- Install LeRobot and calibrate both arms.
- Mount at least one camera, ideally two.
- Teleoperate and record dozens of demonstrations.
- Train a policy and run it on the follower.
Do you need to 3D print the parts yourself, or can you buy the arm assembled? Either works. Seeed and other resellers sell kits with and without printed parts, and some sell assembled arms for more. Printing yourself saves money but costs you an afternoon of tuning tolerances.
Is the SO-101 worth it, or should you just get a myCobot? Decide by goal. For collecting AI training data, the SO-101 wins on price and on community support, because LeRobot’s tutorials, datasets and pretrained policies assume its hardware. For a classroom that needs a ready-made arm with Blockly and a vendor to call, a myCobot 280 is easier.
Tier 4: Desktop educational arms ($649โ$1,014): mechArm, myCobot, DOBOT Magician Lite
These arms arrive assembled, with smart servos, a vendor app, Blockly, a Python SDK and usually ROS support. Elephant Robotics makes the widest range: the mechArm 270, myCobot 280, myCobot 320, myCobot Pro 630, the 4-axis ultraArm P340, the myPalletizer 260 and the 7-axis myArm 300.
On the Elephant Robotics store, the mechArm M5Stack lists at $649, the mechArm Pi at $799 and the mechArm All-in-one Suite at $1,014 (regular price $1,128.96). The AI Kit vision add-on is $499 extra. DOBOT’s 4-axis Magician Lite lists at $995 at the US education resellers Stemfinity and School Specialty.
| Model | Maker | DoF | Payload | Listed price (USD) | Programming | Where to buy (US) |
|---|---|---|---|---|---|---|
| mechArm 270 (M5Stack) | Elephant Robotics | 6 | ~250 g | $649 | Blockly, Python, ROS | Elephant Robotics store |
| mechArm 270 (Pi) | Elephant Robotics | 6 | ~250 g | $799 | Blockly, Python, ROS | Elephant Robotics store |
| mechArm All-in-one Suite | Elephant Robotics | 6 | ~250 g | $1,014 | Blockly, Python, ROS | Elephant Robotics store |
| myCobot 280 (M5) | Elephant Robotics | 6 | 250 g | ~$649 | Blockly, Python, ROS/ROS 2 | Elephant Robotics store |
| Magician Lite | DOBOT | 4 | ~250 g | $995 | Blockly, Python | Stemfinity, School Specialty |
What you give up: payload and stiffness. A 250 g rating is measured under ideal conditions, and the arm’s own gripper uses up part of it. Repeatability on a myCobot 280 is quoted at around half a millimetre. That is fine for stacking blocks and teaching MoveIt, but not for pressing a pin into a tight hole.
Tier 5: Research and desktop industrial arms ($2,999โ$8,995)
Tier 5 is where you get real encoders, metal structures and payloads of up to 1.5 kg on a benchtop. It mixes lab arms built for data collection with small industrial arms for electronics benches.
UFactory’s official Lite 6 page starts at $2,999 for a 600 g payload. US resellers usually list the bare arm at about $3,500, with gripper kits from about $3,600 to $4,500. DOBOT’s MG400, a 4-axis desktop arm, lists at $2,999 at Trossen Robotics, which showed it out of stock when checked. The DOBOT M1 Pro SCARA carries 1.5 kg and lists at about $5,500 at Openelab’s US store. The end effector, guarding and fixtures cost extra. Niryo’s 6-axis Ned2 (300 g) starts at $8,995 at RobotLAB.
| Model | Maker | Type | Payload | Listed price (USD) | Software | Where to buy (US) |
|---|---|---|---|---|---|---|
| Lite 6 | UFactory | 6-axis arm | 600 g | From $2,999 official; ~$3,500 resellers | Python, ROS/ROS 2 | UFactory store, US resellers |
| MG400 | DOBOT | 4-axis desktop | 750 g | $2,999 | DOBOT Studio, Python | Trossen Robotics, Top3dshop |
| M1 Pro | DOBOT | SCARA | 1.5 kg | ~$5,500 | DOBOT Studio, Python | Openelab, DOBOT resellers |
| Ned2 | Niryo | 6-axis | 300 g | From $8,995 | Blockly, Python, ROS | RobotLAB |
| Interbotix and WidowX AI | Trossen Robotics | 4โ6 DoF | 50โ750 g (Interbotix) | Varies by model | ROS 2, Python | Trossen Robotics store |
Lab platforms sell on their data pipelines and ROS 2 drivers as much as on mechanics. Dynamixel-based arms from ROBOTIS and Trossen’s Interbotix line report position, velocity, current and temperature on every joint, and that telemetry turns a robot into a research instrument. Elephant Robotics’ myArm M&C teleoperation pair and AgileX’s OPEN NERO bimanual platform serve the same data-collection market. Generation Robots lists OPEN NERO at โฌ8,280 including VAT, which is a European reference price rather than a US quote.
Rethink Robotics’ Baxter, a two-armed research robot from the early 2010s, still appears in some catalogues. Treat it as legacy hardware and check who supplies parts and software updates before you pay.
Tier 6: Budget industrial cobots ($6,999โ$9,500): Fairino, DOBOT Nova, UFactory xArm
This is where industrial payload begins: 2 to 5 kg, controller included, with the safety functions a cobot application needs. The US reseller Top3dshop lists the 5 kg Fairino FR5 at $6,999, including the control box and an e-stop button box. Fairino claims ยฑ0.02 mm repeatability and 922 mm reach. The Fairino family runs from the FR3 (3 kg) to the FR10 (10 kg) and above.
The DOBOT Nova 2, a 6-axis cobot with a 2 kg payload and about 625 mm reach in most listings, is $8,850 on Dobot’s US store and at Digi-Key. Other resellers list it as high as $12,790, so compare offers before you buy. UFactory’s 5 kg xArm 6 sits around $9,500 in buyer guides, and the xArm 5 Lite is UFactory’s cheaper 5-axis option.
| Model | Payload | Listed price (USD) | Notes | Where to buy (US) |
|---|---|---|---|---|
| Fairino FR5 | 5 kg | $6,999 | Includes controller and e-stop box | Top3dshop |
| DOBOT Nova 2 | 2 kg | $8,850 | 6-axis, 1-year limited warranty | Dobot US store, Digi-Key |
| UFactory xArm 6 | 5 kg | ~$9,500 | Python SDK, ROS support | UFactory store, resellers |
Compared with premium cobots, you give up ecosystem depth, local service and in some cases the third-party safety certification documents your integrator will ask for. Ask for those documents before you order.
Tier 7: Premium cobots ($38,000โ$49,000): Universal Robots and peers
Universal Robots doesn’t publish list prices. Vention lists the UR7e at $38,363, and Automation Distribution lists it at $43,681โ$48,682. The UR7e replaces the long-running UR5e and carries 7.5 kg. UR’s datasheet gives 850 mm reach, and a separate UR7e 920 variant reaches further.
The extra money buys the industry’s largest accessory ecosystem (UR+), integrators in almost every region, documented safety functions and a resale market. Doosan Robotics competes here with its M-series (3โ5 kg), A-series (6โ15 kg) and H-series (6โ20 kg), sold on quote through integrators. The Franka Research 3 is the research counterpart. It has 7 DoF, a 3 kg payload, 855 mm reach and joint torque sensing that labs use for force-control work, and it is sold on quote.
| Model | Payload | Listed price (USD) | Notes | Where to buy (US) |
|---|---|---|---|---|
| Universal Robots UR7e | 7.5 kg | $38,363โ$48,682 | UR5e successor, UR+ ecosystem | Vention, Automation Distribution |
| Doosan M/A/H series | 3โ20 kg | Quote only | Wide payload range | Doosan integrators |
| Franka Research 3 | 3 kg | Quote only | 7 DoF, torque sensing, ROS 2 | Franka Robotics, distributors |
The counter-argument deserves a fair hearing. A Fairino FR5 costs less than a fifth of a UR7e, and for one simple machine-tending job the cheaper arm often pays back faster. The UR premium earns its keep when you need a local integrator, certified safety paperwork for an auditor, or many cells that share spare parts and programs.
The listed price is not the system price: grippers, cameras, mounts, safety, integration
The arm is often less than half the cost of a working robot cell. Before it does anything useful, a cobot needs these extras:
- End effector. An electric gripper or vacuum system. The Lite 6 bundle price rises by several hundred dollars with a gripper, and industrial grippers cost far more.
- Vision. A camera, lighting and calibration if parts arrive in random positions.
- Mounting. A rigid pedestal or a machined plate. A 5 kg cobot on a flimsy table will vibrate, and its precision will suffer.
- Safety. A risk assessment, and sometimes scanners, light curtains or guarding. A sharp tool can call for these even on a cobot.
- Integration. Programming, fixtures, connecting to the machine you’re tending, testing and staff training.
- Spares and downtime. Spare servos for desktop arms, a service contract for cobots.
A worked example: a small shop wants a Fairino FR5 to load a CNC lathe. The arm is $6,999. Add a gripper, a pedestal, part-presentation fixtures, I/O wiring to the lathe and a risk assessment, and the project can easily reach two or three times the arm’s price before the first part runs. Budget for that on day one.
Specs that matter, and how spec sheets mislead you
Five specs decide whether an arm can do your job: payload at your working reach, reach, repeatability, speed, and the controller and software. A spec sheet can mislead you on all five, mostly by quoting the best case without saying so.
| Spec | What it means | How it misleads | Question to ask the vendor |
|---|---|---|---|
| Payload | Mass the arm can carry, including the tool | Rated near the base or at slow speed | “What is the payload at full reach and full speed?” |
| Reach | Distance from base axis to wrist or flange | The edges of the workspace are where accuracy is worst | “What’s the usable workspace for my tool’s orientation?” |
| Repeatability | How closely it returns to a taught point | Often quoted where accuracy matters more | “What is the absolute positioning accuracy?” |
| Speed | Maximum joint or tool speed | Not achievable with payload or near people | “What’s the cycle time for my move with my payload?” |
| Actuators | PWM servo, smart bus servo or geared industrial joint | Hidden behind marketing names | “What feedback does each joint report?” |
| Software | SDK, ROS 2 driver, MoveIt config | “ROS supported” can mean an unmaintained repo | “Which ROS 2 distribution, and when was it last updated?” |
Actuators set the ceiling for everything else. From the computer’s point of view, PWM hobby servos are open-loop. Smart bus servos (Dynamixel, Feetech STS) close the loop inside the servo and report their state over a serial bus. Industrial joints with harmonic gearing and dual encoders give stiffness and precision that plastic gears can’t match.
Repeatability is not accuracy
Repeatability measures how tightly the arm returns to the same taught point. Accuracy measures how close it gets to a point you command in coordinates. A UR7e repeats to within about ยฑ0.03 mm, but its absolute accuracy without calibration is noticeably worse, because small errors in link lengths and joint offsets add up.
This affects how you program. Teach-by-demonstration, where you jog to a point and save it, depends on repeatability. Programming from a CAD model or from a camera’s coordinates depends on accuracy. Hobby and desktop arms often quote only repeatability, and some quote neither across the full workspace.
Engineer's note
Ask for both repeatability and accuracy, and ask where in the workspace they were measured. A figure taken at the centre of the workspace with no load says little about the far edge with a gripper attached.
Why a “500 g payload” arm can’t lift 500 g at full reach
The load on a joint is a torque: mass times gravity times horizontal distance. Hold 500 g at 0.3 m from the shoulder and that joint needs about 1.5 Nยทm (roughly 15 kgยทcm) for the payload alone. It also carries the weight of every link and servo beyond it.
A cheap metal-gear hobby servo is typically rated around 10 kgยทcm, and that is stall torque, the point where the motor stops. Usable continuous torque is a fraction of that. So a “500 g” kit arm might manage 500 g tucked close to its base and much less at full extension. Ask for more and it will droop, jitter or overheat.
That also explains why a servo arm jitters while it’s just holding position. There are three usual causes. The load sits near the servo’s deadband, so it hunts back and forth. The power supply sags when several servos draw current at once. Or the signal ground is poorly shared with a noisy supply. The fixes are a dedicated 5โ6 V supply rated for the combined stall current, shorter lever arms, lighter payloads, or smart servos with tunable PID gains.
Engineer's note
Before you compare products, write down the exact task, the heaviest object including the gripper, and the farthest point the arm must reach. Then rule out every arm whose rated payload at that reach is too low.
Best robotic arm for your goal
Choose by goal first and budget second. The same $700 is well spent on a myCobot for a ROS 2 course and wasted on one for factory packing. These picks follow four weighted criteria, drawn from manufacturer specs, US listed prices and community documentation: fit to the task (payload and reach), software openness (SDK, ROS 2, LeRobot), total cost including accessories, and support (spares, vendor activity, integrators). Fit to the task carries the most weight, because an arm that can’t do the job is poor value at any price.
| Goal | Tier | Example | Listed price (USD) | Why |
|---|---|---|---|---|
| Learn Arduino and servo control | 2 | Hiwonder or Waveshare kits | $100โ$400 | Cheap, hackable, cheap to break |
| Collect imitation-learning data | 3 | SO-101 leader and follower | $220โ$240 kit | Native LeRobot support, big community |
| Learn ROS 2 and MoveIt | 4 | myCobot 280, mechArm 270 | $649โ$799 | URDF and ROS packages available |
| Teach a class | 4 | DOBOT Magician Lite, myCobot 280 | $649โ$995 | Blockly, safe on a desk, vendor curriculum |
| Electronics bench or SCARA picking | 5 | DOBOT MG400, DOBOT M1 Pro | $2,999โ~$5,500 | Fast, rigid, compact |
| University research lab | 5 or 7 | Lite 6, WidowX AI, Franka Research 3 | $2,999 to quote | Torque sensing, ROS 2, data tools |
| Automate machine tending or packing | 6 or 7 | Fairino FR5, DOBOT Nova 2, UR7e | $6,999โ$48,682 | Real payload, safety functions |
Best for kids and STEM
For ages 8โ12, a hydraulic kit like the Teach Tech Hydrobot teaches mechanics without screens. Once kids are coding, an arm with Blockly support, such as the DOBOT Magician Lite or the myCobot 280, lets them move from drag-and-drop blocks to Python on the same hardware. That continuity is worth more than a better spec sheet.
Best robotic arm kit for adults
For electronics, start with a servo arm under $300 and expect to fight jitter, because that struggle is the lesson. If you want a cheap arm that behaves predictably, a kit built on smart bus servos is the better buy. The SO-101 gives you position feedback and an active community at hobby-kit prices, which makes it the strongest open-source kit in 2026.
Best for AI and embodied research
An SO-101 pair is the entry point. ROBOTIS arms and the Trossen WidowX AI are the step up. Research groups that need force control or published benchmarks use the Franka Research 3. To see how a robot learns from demonstrations, the leader/follower pair is the cheapest honest way to try it.
Best for a small business
If your parts weigh more than about 1 kg, desktop arms are out, and a 5 kg cobot is the realistic floor. Fairino, DOBOT and UFactory lead on price. Universal Robots costs more, but its integrator network and accessories often shorten deployment. Run a pilot on one task, measure the cycle time, and only then scale up.
Where to buy a robotic arm in the US (and buying traps)
Buy desktop arms directly from the maker or an authorised distributor, and buy cobots through an integrator or authorised reseller. Amazon and other marketplaces suit cheap kits but are riskier for anything that needs firmware updates, spare parts or warranty service.
| Channel | US examples | Good for | What changes |
|---|---|---|---|
| Maker’s store | Elephant Robotics, UFactory, Dobot US, Seeed Studio | Desktop arms, kits, some cobots | Current versions, direct support, sometimes slow shipping |
| Authorised distributor | Top3dshop, Trossen Robotics, RobotLAB, Digi-Key, Stemfinity | Desktop and research arms, cobots | Local warranty, bundles, education pricing, sometimes tariffs included |
| Online cobot marketplace | Vention, Automation Distribution | UR and other cobots | Published prices for brands that don’t list them |
| Integrator | Regional UR, Doosan and FANUC partners | Cobots and complete cells | Higher price, but includes risk assessment and commissioning |
| Amazon and marketplaces | Hiwonder, Waveshare, unbranded kits | Toy and hobby kits | Fast delivery, mixed listings, uncertain spare parts |
The traps to watch for:
- EUR with VAT vs USD before tax. Generation Robots in Europe lists the DOBOT Magician Lite at โฌ780 including VAT (โฌ650 excluding). US resellers list it at $995 before sales tax. European listings work as a reference only, so compare like with like.
- Wide reseller spreads. The DOBOT Nova 2 is $8,850 on Dobot’s US store but up to $12,790 elsewhere. Always check the maker’s own price.
- Retired models still on sale. The UR7e has replaced the UR5e, but older listings remain. Old stock can be a bargain if support continues, so ask how long it will.
- Tariffs and stock. Some US resellers now include tariffs in listed prices, and some UFactory and DOBOT listings show backorders. Confirm the delivery date before you pay.
- Sold-out or ghost kit listings. Popular open-source kits sell out after launches, and third-party copies fill the gap. Check that the servos match the official bill of materials.
- Spare servos and warranty. A burned-out servo shouldn’t end a desktop-arm project. Check that individual servos are sold separately.
What a robotic arm actually is
A robotic arm is a serial chain. A fixed base carries a series of rigid links, and motorised joints connect them. A wrist at the end sets the tool’s angle, and the end effector does the work. If you know every joint angle, you can calculate where the tool is (forward kinematics). Working backwards from a target pose to the joint angles is inverse kinematics.
The base is bolted to a table, a pedestal or a mobile platform, and its stiffness matters more than most beginners expect. A desktop arm clamped to a wobbly desk inherits the wobble.
Joints come in two kinds:
- Revolute joints rotate, like your elbow. Nearly every articulated arm is built from them.
- Prismatic joints slide in a straight line. Cartesian systems are built from them, and a SCARA robot uses one for its vertical axis.
Each joint has an actuator, usually an electric motor and gearbox, plus some way of sensing position. Cheap hobby servos sense position with a potentiometer. Smart bus servos such as ROBOTIS Dynamixel or the Feetech units in the SO-101 report position digitally. Industrial joints use high-resolution encoders and strain-wave (harmonic) gearing. That single difference explains most of the precision gap between tiers.
Degrees of freedom: 4, 5, 6 or 7?
Six degrees of freedom is the general-purpose baseline. Three joints place the tool anywhere in the workspace, and three wrist joints point it in any direction. With fewer axes, some combinations of position and angle are out of reach. With seven, the arm can reach the same pose in many ways.
| DoF | What you can do | What you cannot do | Example |
|---|---|---|---|
| 4 | Pick-and-place from above with tool rotation | Tilt the tool sideways | DOBOT Magician Lite, DOBOT MG400, myPalletizer 260 |
| 5 | Reach and tilt for most tabletop picks | Some orientations at some positions | Many Interbotix and hobby arms, xArm 5 Lite |
| 6 | Any position and orientation within reach | Avoid obstacles while holding a pose | myCobot 280, UR7e, xArm 6, Fairino FR5, DOBOT Nova 2 |
| 7 | Same pose with different elbow positions (redundancy) | Nothing extra in pose terms, but control is more complex | Franka Research 3, myArm 300 |
Seven axes earn their cost in cluttered spaces. The Franka Research 3 can keep its tool still while it swings its elbow around an obstacle, much as a human arm does. The catch is that the inverse kinematics have infinitely many solutions, so planning software has to choose among them.
Many hobby listings count the gripper as an axis. A “6-DoF” kit with a gripper often has five positioning joints. Check the joint list, not the headline.
End effectors: grippers, suction cups, cameras
The end effector is whatever the arm carries, and its weight counts against the payload. Common types:
- Parallel two-finger grippers: the default for boxes, blocks and machined parts. Electric on desktop arms, electric or pneumatic on cobots.
- Adaptive or soft grippers: fingers that wrap around irregular shapes such as fruit.
- Vacuum suction cups: the fastest option for flat, sealed surfaces like cardboard, sheet metal and glass. They fail on porous or mesh surfaces.
- Cameras: a wrist camera (“eye-in-hand”) or a fixed overhead camera (“eye-to-hand”) for vision-guided picking. Elephant Robotics’ $499 AI Kit is built on this hardware.
- Process tools: welding torches, screwdrivers, dispensers and pens.
An “object manipulator” usually means the arm and gripper together. In engineering terms, the manipulator is the arm and the gripper is its tool, and you size them as one system.
Types of robotic arms and what each is good at
There are six practical arm types, and each trades flexibility against speed, stiffness or cost. Articulated arms dominate general work. SCARA and delta robots win at fast flat picking. Cartesian gantries win on large, precise rectangular workspaces.
| Type | How it moves | Strengths | Where it fails | Typical jobs | Named examples |
|---|---|---|---|---|---|
| Articulated (6-axis) | Chain of revolute joints | Reaches around objects, very flexible | Stiffness drops at full reach, complex kinematics | Welding, machine tending, assembly | Fairino FR5, UFactory xArm 6 |
| SCARA | Two horizontal rotary joints plus a vertical slide | Fast, rigid vertically, cheap per cycle | Can’t tilt the tool | Electronics assembly, screwdriving | DOBOT M1 Pro |
| Delta (parallel) | Three arms joined at one platform | Extremely fast light picks | Tiny payloads, small workspace | Food packing, sorting | ABB IRB 360 FlexPicker |
| Cartesian / gantry | Linear X, Y, Z axes | Huge, precise, easy to program | Large footprint, can’t reach around things | CNC loading, 3D printers, palletising | CNC loaders, 3D printer motion systems |
| Collaborative (cobot) | Usually 6-axis articulated with force limiting | Fast deployment, can work near people | Speed limited when people are near | Packing, tending, screwdriving | Universal Robots UR7e, DOBOT Nova 2 |
| Desktop / educational | Small articulated or 4-axis arms | Cheap, safe on a desk, easy to program | Low payload and stiffness | Teaching, prototyping, AI research | myCobot 280, DOBOT MG400, ultraArm P340 |
For most buyers, “robotic arm robot” means the articulated 6-axis type. It’s what you see on car body lines, and it’s what nearly every cobot copies at a smaller scale.
Cobot vs industrial arm: what “collaborative” really means
A cobot is an arm built to run safely near people without a fence. It gets there by limiting force and power, monitoring speed and separation, or using hand guiding. “Collaborative” describes a set of safety functions and an application. It doesn’t certify that the robot is safe.
A UR7e can pack cartons beside a worker with no danger at all. Bolt a knife or a hot glue gun to it and the same robot becomes dangerous, because the tool changes the risk. That’s why safety standards assess the whole application: robot, tool, workpiece and layout. A traditional industrial arm, meanwhile, can work collaboratively if area scanners slow or stop it when someone approaches.
With a cobot you give up speed. Force limits and lighter structures mean a cobot usually cycles more slowly than a caged industrial arm with the same payload. You gain floor space, faster setup, and programming by hand guiding or tablet instead of a specialist’s teach pendant.
Did you know?
According to A3, cobots made up 28.6% of all robots ordered in North America in Q4 2025 but only 10.7% of full-year order revenue. They sell in volume at much lower unit prices than traditional industrial arms.
Software: what you’ll actually program it with
Most arms offer some mix of Blockly for beginners, a Python SDK, ROS or ROS 2 drivers, and MoveIt for motion planning. LeRobot is the newer framework for learned policies. An open, documented API outlasts any single spec. An arm locked to a closed app is a dead end once you outgrow the app.
The common layers:
- Blockly and vendor apps: drag-and-drop sequences. Elephant Robotics, DOBOT and Niryo all ship one.
- Python SDKs: Elephant Robotics’ pymycobot, UFactory’s xArm Python SDK and DOBOT’s APIs let you script motion, I/O and grippers.
- ROS 2: the standard robotics middleware. Look for drivers built for a current long-term-support distribution such as Humble or Jazzy.
- MoveIt: the ROS motion-planning framework. It handles inverse kinematics, collision checking and trajectory planning from the arm’s URDF model.
- LeRobot: Hugging Face’s Python library for recording datasets, training policies and running them on low-cost arms.
- rosbridge: exposes ROS topics over WebSockets so a browser dashboard can control or monitor the arm.
Once a driver is running, a quick check confirms the arm is publishing joint state before you invest time in MoveIt:
ros2 node list
ros2 topic echo /joint_states --once
# Generate a MoveIt config from the arm's URDF if the vendor doesn't ship one
ros2 launch moveit_setup_assistant setup_assistant.launch.py
Industrial controllers run real-time loops on dedicated hardware, which is why their motion stays jitter-free under load. Teams that simulate a cell before buying hardware rely on published URDF and MoveIt files to build a digital twin.
Engineer's note
Before buying, check for a maintained Python SDK, ROS 2 support and published URDF and MoveIt configs. Look at the date of the last commit on the vendor’s repository, not the marketing page.
Safety and standards: ISO 10218:2025 in plain English
ISO 10218 is the international safety standard for industrial robots, and the 2025 revision is the biggest update since 2011. It folds the collaborative-robot guidance from ISO/TS 15066 into the main standard, separates manufacturer duties from integrator duties, and adds cybersecurity requirements.
A3 announced the revised standard in February 2025 after about eight years of committee work. The key changes:
| Change | ISO 10218:2011 + ISO/TS 15066 | ISO 10218:2025 |
|---|---|---|
| Collaborative requirements | Separate technical specification (ISO/TS 15066) | Built into the standard’s collaborative-application section |
| Who it addresses | Parts 1 and 2, with blurrier duties | Part 1: robot manufacturers. Part 2: integrators and applications |
| Functional safety | Single set of requirements | Robot classes that scale the requirements |
| Cybersecurity | Not addressed in depth | Requirements included, since cyber threats can affect safety |
ISO 10218-1:2025 replaces the 2011 edition. Its collaborative-application content is the former ISO/TS 15066 material, including the body-region force and pressure limits used to validate power-and-force-limited applications. In the US, A3 maintains the national adoption as ANSI/A3 R15.06, which follows the ISO text.
For a buyer, the point is simple: a cobot still needs a risk assessment. Part 2 puts that responsibility on whoever integrates the application, and if you install the robot yourself, that is often you. Desktop educational arms usually fall outside these industrial standards. Still, treat their pinch points with respect around children.
Reality check
No robot is safe without guarding just because it is sold as a cobot. Safety is proven for each application, including the tool, the part and the layout, through a documented risk assessment.
The robot arm market in brief (2025 data)
Demand for robot arms hit a record in 2025. The IFR’s World Robotics 2026 report (September 24, 2026) counts 603,307 industrial robot installations, up about 11%, and an operational stock of about 5 million. China installed about 354,000 units, 59% of the global total. The US passed Japan for second place with about 38,400 units, up 12%, while Japan fell 19% to 36,219. The IFR forecasts 655,000 installations in 2026. That is a projection, not a measurement.
A3 counts North American orders rather than installations, so its figures don’t line up with the IFR’s. Companies ordered 36,766 robots worth $2.25 billion in 2025, and cobots made up 19.6% of the units. Food, electronics and life-sciences buyers out-ordered automotive, a shift that favours smaller, flexible arms over the welding giants from FANUC and ABB.
Robotic arms for humans: prosthetic and bionic arms
A robotic arm “for humans” is a prosthetic or bionic arm, a medical device fitted by a certified prosthetist and usually billed to an insurer. Most powered versions use myoelectric control: sensors on the residual limb read muscle signals and drive motorised fingers, wrists or elbows.
How fitting works
You don’t buy a bionic arm off a shelf. In the US, the process usually runs through a physician’s prescription and a prosthetist. Look for the CP (Certified Prosthetist) credential from the American Board for Certification in Orthotics, Prosthetics & Pedorthics (ABC). A typical path:
- Clinical evaluation and a prescription from a physician.
- Assessment by the prosthetist, including testing which muscle sites give usable signals.
- Casting or scanning, then one or more test sockets.
- Fitting of the definitive prosthesis and the powered components.
- Occupational therapy to learn the controls, followed by adjustments and maintenance.
What it costs and what insurance covers
The device is only part of the bill, which also covers the socket, fitting, therapy and follow-up. A US Department of Veterans Affairs cost study, cited by CostHelper and now dated, put myoelectric prostheses at $18,703 for partial hand loss, $20,329 at mid-forearm, $59,664 at mid-upper-arm and $61,655 at the shoulder. In January 2025, KFF Health News reported on a patient who said her myoelectric hand would cost just over $46,000 without insurance.
Medicare covers prosthetic limbs it deems medically necessary at 80% of the approved amount, which leaves you 20% coinsurance, according to KFF Health News. Private coverage “varies tremendously,” Ashlie White of the Amputee Coalition told the outlet, and most people pay something toward the device. Insurers can cap benefits or restrict which device types they approve. About half of US states have prosthetic “insurance fairness” laws, but those laws bind only state-regulated plans, and more than half of privately insured people are in plans that state law doesn’t govern.
The best-known price point comes from television. On Shark Tank Season 15, Episode 16 (aired February 2024), Psyonic founder Dr. Aadeel Akhtar said the Ability Hand sold for $15,500 and cost about $1,800 to make, according to the episode transcript. San Diego Magazine later reported that each hand costs between $15,000 and $20,000. Ask Psyonic for a current quote through your prosthetic provider rather than relying on the TV figure.
If you’re exploring this for yourself or someone close to you, start with a certified prosthetist. They can tell you which devices suit your limb and what your insurer covers.
Did you know?
A Government Accountability Office report cited by KFF Health News found that only 30% of Medicare beneficiaries who lost a limb in 2016 received a prosthesis within the following three years.
The Shark Tank robotic arms: what happened
Three robotics companies made on-air deals on Shark Tank Season 15: Psyonic, Chefee and RoboBurger. Only Psyonic’s deal is reported closed, and that rests on third-party data.
| Company | Product | On-air deal | Status, per reporting |
|---|---|---|---|
| Psyonic | Ability Hand bionic hand | $1M from Kevin O’Leary, Lori Greiner and Daymond John | Reported closed (appears in their PitchBook portfolios) |
| Chefee | Robotic home kitchen | $500,000 for 15% from Kevin O’Leary | Closing unconfirmed |
| RoboBurger | Automated burger vending robot | $1.5M total from Kevin O’Leary and Michael Rubin | Closing unclear |
Psyonic. Akhtar asked for $1 million, and three sharks invested together. Reports disagree on the equity, citing either 2% or 6%. BGR reports the deal closed and that Psyonic is targeting capacity of 10,000 hands a year.
Chefee. Founder Assaf Pashut accepted O’Leary’s offer of $500,000 for 15% in an episode that aired March 8, 2024. Several outlets report the deal wasn’t finalised during due diligence.
RoboBurger. O’Leary and Rubin each put in $750,000 on Season 15, Episode 21. Whether the deal closed hasn’t been publicly confirmed.
FAQ
How much does a robotic arm cost?
It depends on the tier. Toy kits cost under $100 and hobby servo arms $100โ$400. An open-source SO-101 is $130โ$240, and desktop educational arms run $649โ$1,014. Research and desktop industrial arms cost $2,999โ$8,995, budget industrial cobots $6,999โ$9,500, and premium cobots such as the UR7e $38,000โ$49,000. A complete cell with integration costs more.
What is the cheapest robotic arm that can do useful work?
For light repetitive desk tasks such as sorting small parts, a 6-axis desktop arm at $649โ$799 is enough. For heavier benchtop work, the UFactory Lite 6 and DOBOT MG400 start at $2,999. For industrial parts up to 5 kg, the Fairino FR5 at $6,999 is the lowest listed US price in this guide.
Where can I buy a robotic arm in the US?
Buy from the maker’s store (Elephant Robotics, UFactory, Dobot US, Seeed Studio) or from an authorised distributor such as Top3dshop, Trossen Robotics, RobotLAB, Digi-Key or Stemfinity. Vention and Automation Distribution publish Universal Robots prices. For a complete cobot cell, go through an integrator.
Are robotic arms on Amazon worth buying?
For toy and hobby kits under $400, yes, as long as you check what feedback the servos give. For anything that needs firmware updates, spare parts or warranty service, buy from the maker or an authorised reseller.
What is the best robotic arm kit for adults?
The SO-101 is the best open-source kit in 2026. It’s cheap, its smart servos report their position, and LeRobot supports it directly. If you’d rather not assemble anything, a myCobot 280 or mechArm 270 works out of the box with Python and ROS support.
How much does a cobot cost compared with an industrial robot arm?
Budget industrial cobots list at $6,999โ$9,500, and premium cobots like the UR7e at $38,000โ$49,000. In A3’s 2025 data, cobots made up 19.6% of North American robot orders by units but only 10.7% by revenue, which reflects their lower average price. Traditional industrial arms also need guarding, which raises the system cost.
How much does a prosthetic or bionic arm cost?
Myoelectric prostheses commonly run from about $20,000 to over $60,000 depending on amputation level, based on a dated VA cost study. Medicare pays 80% of the approved amount for covered devices. Private coverage varies widely, so a certified prosthetist and your insurer are the only reliable sources for your actual cost.
What happened to the robotic arm on Shark Tank?
Psyonic, maker of the Ability Hand bionic hand, received $1 million from Kevin O’Leary, Lori Greiner and Daymond John on Season 15, and the deal reportedly closed. Chefee and RoboBurger also made on-air deals, but neither has been confirmed as closed.
How much does an AI girlfriend robot cost?
That search refers to companion or humanoid products, not robotic arms, and none of the arms in this guide belong to that category. If what interests you is AI on real robot hardware, an SO-101 leader/follower kit at $220โ$240 is the cheapest way to train a learned behaviour on a physical robot.
How many degrees of freedom does a robotic arm need?
Six for general-purpose work, since that lets the tool reach any position and orientation within range. Four is enough for top-down pick-and-place, as on the DOBOT MG400. Seven adds redundancy for working around obstacles, as on the Franka Research 3.
Can a robotic arm be programmed with Python or ROS 2?
Most current arms support both. Elephant Robotics, UFactory, DOBOT and Niryo ship Python SDKs, and most offer ROS or ROS 2 drivers. LeRobot is Python-native. Before buying, check that the ROS 2 driver targets a current distribution and has been updated recently.
Is a collaborative robot safe to use without guarding?
Only if a risk assessment of the whole application says so. Under ISO 10218-2:2025, the integrator validates the tool, the workpiece, the speeds and the forces. A cobot carrying a sharp tool can still need scanners or guarding.
What is the difference between repeatability and accuracy?
Repeatability is how closely the arm returns to a point it was taught. Accuracy is how close it gets to a point you specify in coordinates. Teach-by-demonstration depends on repeatability. CAD-based and vision-based programming depend on accuracy, which is usually the weaker figure.
Is there anything better than a hobby arm without spending cobot money?
Yes. Arms built on smart bus servos, such as the SO-101, ROBOTIS-based arms and the myCobot 280, report their joint positions and hold them far better than PWM servo kits. The next step up is tier 5, where the UFactory Lite 6 and DOBOT MG400 start at $2,999.
Before you pay, check one more thing: ask the seller for the price of a replacement joint or servo and how long they expect to stock it. A seller who answers that clearly is the one who will still support the arm in three years, and that support is worth more than a small discount at checkout.
Credits and sources
By RWT Robotics Editorial Team
- Facts were checked against the sources below before publication.
- Published
Checking and sources
- Researched from 8 sources, listed below.
- Facts are checked against those sources; anything we cannot confirm is corrected, labelled or removed.
Sources (8)
- IFR: Five million robots now operate in factories globally (ifr.org)
- IFR World Robotics 2026: 603k installs, five million stock (insidedeeptech.com)
- IFR press release World Robotics 2026: China (ifr.org)
- IFR press release World Robotics 2026: USA (ifr.org)
- The Robot Report: 5 million robots now working in factories worldwide (therobotreport.com)
- A3: Robot orders grow 6.6% in 2025 (automate.org)
- The Robot Report: North American robot orders rise 6.6% in 2025 (therobotreport.com)
- SDCExec: Robot demand from F&B, semiconductors, life sciences outpaced automotive (sdcexec.com)









