A PUMA arm is a 6-degree-of-freedom serial articulated robot with three large positioning joints (waist, shoulder, elbow) followed by a 3-axis spherical wrist. Each joint rotates independently, and the kinematic chain stacks those rotations to place the tool centre point anywhere inside a roughly spherical workspace at any orientation. The design solves the problem of getting a tool to an arbitrary 3D pose with a compact, anthropomorphic footprint. Unimation introduced the PUMA 560 in 1978, and that geometry still defines how most industrial 6-axis robots are built today.
PUMA Arm Interactive Calculator
Vary the waist, shoulder, elbow, and link lengths to see the wrist-center position of a PUMA-style serial robot.
Equation Used
This calculator uses the first three PUMA joints as a serial positioning chain. The shoulder and elbow create a planar wrist-center radius and height, then the waist angle rotates that radius into X and Y. The spherical wrist is assumed to intersect at that wrist center, so orientation is decoupled from position.
- Joints 1-3 set wrist-center position only.
- Spherical wrist axes intersect at the wrist center.
- Link offsets and tool length are ignored.
- Angles are ideal commanded joint angles.
How the PUMA Arm Actually Works
PUMA stands for Programmable Universal Machine for Assembly. The arm is a serial chain — joint 1 is a vertical waist rotation, joint 2 lifts the shoulder, joint 3 bends the elbow, then joints 4-5-6 form a spherical wrist where all three axes intersect at a single point. That intersection is the trick. When the wrist axes meet at one point, the inverse kinematics decouple — you solve position with the first three joints and orientation with the last three, separately. Without that decoupling you'd need a numerical solver running every cycle, and cycle times go up.
Why this geometry? Because it mirrors a human arm well enough to reach into the same kinds of spaces a human worker would, but with a smaller footprint than a Cartesian gantry covering the same volume. The tradeoff is stiffness — a serial chain accumulates compliance at every joint, so a PUMA arm flexes more under load than a parallel robot like a delta or a Stewart platform. Repeatability on a PUMA 560 is around ±0.1 mm at the flange, but absolute accuracy is worse, typically ±1-2 mm before calibration, because every link length and joint offset stacks tolerance.
Things go wrong when the kinematic parameters drift. If joint 4 and joint 6 line up — wrist singularity — the controller loses one degree of freedom and either freezes or whips the wrist around at high speed to escape. If the harmonic drives in the wrist wear, backlash shows up as orientation chatter at the tool, usually visible first on circular paths where the wrist has to reverse direction. And if the Denavit-Hartenberg parameters in the controller don't match the as-built arm to within 0.05 mm and 0.01°, you'll see position error that grows as the arm extends — a classic sign the link-length calibration is off.
Key Components
- Waist (Joint 1): Rotates the entire arm around a vertical axis at the base. On a PUMA 560 this joint covers ±160° at up to 100°/s. The base casting and crossed-roller bearing here carry the full overturning moment of the arm — bearing preload must hold to within 5 µm radial play or the tool centre point wobbles measurably.
- Shoulder (Joint 2): Pitches the upper arm up and down. This is the most heavily loaded joint because it lifts the entire arm plus payload against gravity. PUMA designs use a counterbalance spring or pneumatic cylinder here so the motor doesn't have to hold static load — without it the gearbox would creep under hold torque.
- Elbow (Joint 3): Bends the forearm relative to the upper arm. Range typically ±135°. Combined with shoulder and waist, joints 1-3 set the position of the wrist centre point — the single point where all three wrist axes intersect.
- Spherical Wrist (Joints 4, 5, 6): Three rotary axes whose centerlines all cross at one point. Joint 4 rolls the forearm, joint 5 pitches the wrist, joint 6 rolls the tool flange. The intersection must hold within 0.1 mm or the closed-form inverse kinematics breaks down and the controller falls back to numerical solutions.
- Harmonic Drive Reducers: Most PUMA wrists use harmonic drive gearheads at 100:1 to 160:1 reduction. They give near-zero backlash when new — under 1 arc-minute — but wear opens that to 3-5 arc-minutes after a few thousand hours of duty cycle, and the wear shows up as orientation error at the tool tip.
- Joint Encoders: Each joint runs an absolute or incremental encoder, typically 17-bit to 20-bit resolution. The controller reads all six positions every 1-4 ms and runs forward kinematics to compute tool pose. Encoder phase error above 1 LSB introduces visible jitter on slow linear moves.
Real-World Applications of the PUMA Arm
The PUMA architecture became the default for any task where a human-scale arm needs to reach into a fixture, orient a tool, and apply moderate force. You'll see it everywhere from automotive assembly to pharmacy automation. The reason is workspace versatility — a 6-DOF articulated manipulator covers a roughly spherical reach envelope and can approach a target from almost any angle, which a SCARA or Cartesian robot can't. The cost is stiffness and absolute accuracy, so PUMA-style arms get used where repeatability matters more than absolute positioning, and where the tool can be guided by vision or force feedback to compensate for residual error.
- Automotive Assembly: General Motors deployed Unimation PUMA 560 arms on Buick assembly lines in the early 1980s for small-parts assembly and instrument cluster installation — one of the first widespread industrial uses of a 6-DOF articulated robot.
- Surgical Robotics: Early stereotactic neurosurgery research at Memorial Medical Center in 1985 used a modified PUMA 200 to position a biopsy needle relative to CT imaging — the precursor to modern surgical robots like the ROSA system.
- Laboratory Automation: Stäubli TX2-60 arms (a direct PUMA-geometry descendant) handle microplate transfers in high-throughput screening cells at companies like Beckman Coulter, where 6-axis dexterity lets one arm service multiple instruments arranged radially.
- Arc Welding: FANUC ArcMate 100iD and Yaskawa AR1440 — both 6-DOF articulated arms following PUMA topology — run MIG welding on agricultural equipment at companies like John Deere, where the wrist orients the torch through compound-curved seams.
- Pharmaceutical Packaging: ABB IRB 1200 arms tend vial-filling lines at contract pharma facilities, picking glass vials from infeed pucks and presenting them to stoppering stations at 60-80 cycles per minute.
- Research and Education: The original PUMA 560 became the standard teaching platform for robot kinematics courses at MIT, Stanford, and Carnegie Mellon throughout the 1980s and 1990s, and its DH parameters are still the textbook example in Craig's Introduction to Robotics.
The Formula Behind the PUMA Arm
The forward kinematics tells you where the tool tip ends up given the six joint angles. For a PUMA-style arm with a spherical wrist, the position of the wrist centre depends only on joints 1, 2, and 3 — that's the decoupling that makes inverse kinematics tractable. At the low end of the typical reach envelope (arm folded, ~25% of full reach), the arm is stiffest and most accurate but the workspace is small. At full extension the arm reaches farthest but stiffness drops by roughly 4× because deflection scales with the cube of effective length, and you sit close to the elbow singularity. The sweet spot is 60-80% of full reach where stiffness, manipulability, and clearance from singularities all stay healthy.
Variables
| Symbol | Meaning | Unit (SI) | Unit (Imperial) |
|---|---|---|---|
| Pwrist | Position of the wrist centre point in base-frame coordinates | m | in |
| θ1 | Waist joint angle (rotation about vertical base axis) | rad | deg |
| θ2 | Shoulder joint angle (pitch of upper arm) | rad | deg |
| θ3 | Elbow joint angle (forearm bend relative to upper arm) | rad | deg |
| a2 | Upper arm link length (shoulder to elbow) | m | in |
| a3 | Forearm link length (elbow to wrist centre) | m | in |
| d1 | Base height offset from floor to shoulder axis | m | in |
Worked Example: PUMA Arm in a battery-cell pick-and-place cell at an EV pack assembler
A lithium-ion pack assembler in Nevada runs a Stäubli TX2-90L (PUMA-geometry, 1.2 m reach) to lift 4680-format cells from a tray and seat them into a module fixture 850 mm away horizontally and 300 mm below the shoulder axis. Link lengths are a2 = 0.500 m, a3 = 0.425 m, d1 = 0.450 m. You need to know where the wrist centre lands at three joint configurations to verify the cell stays clear of the fixture wall.
Given
- a2 = 0.500 m
- a3 = 0.425 m
- d1 = 0.450 m
- θ1 = 0 deg (arm pointing along +X)
- θ2 nominal = -20 deg
- θ3 nominal = -40 deg
Solution
Step 1 — at nominal pose (θ2 = -20°, θ3 = -40°), compute the horizontal reach of the wrist centre:
Step 2 — compute the vertical position of the wrist centre at nominal:
That puts the wrist 683 mm out and 89 mm below the shoulder axis — well inside the workspace, comfortably clear of both elbow and shoulder singularities. Manipulability is high here, joint torques are moderate, and stiffness is good because the arm isn't fully extended.
Step 3 — at the low end of the typical operating range, arm folded tighter (θ2 = -10°, θ3 = -90°):
Reach drops to 418 mm. The arm is stiff and accurate here — deflection under a 5 kg cell payload is sub-0.1 mm - but the wrist sits too close to the base to reach the fixture at 850 mm. You'd hit the cell tray with the elbow.
Step 4 — at the high end, arm nearly straight (θ2 = -25°, θ3 = -10°):
Now you're at 801 mm horizontal reach — close to the 850 mm target — but the elbow angle is shallow, you're approaching the elbow singularity (θ3 → 0), and joint 2 is carrying a long moment arm. Stiffness here is roughly a quarter of the nominal pose, and a 5 kg cell can deflect the tool centre point by 0.4-0.5 mm under acceleration — enough to miss the seating chamfer on a tight battery module.
Result
Nominal wrist position lands at (0. 683, 0, -0.089) m relative to the shoulder. That's the configuration you want for cycle-time and accuracy — arm bent enough to stay stiff, far enough from any singularity that the controller solves cleanly. The low-end folded pose gives you 418 mm reach with excellent stiffness but can't hit the target; the high-end extended pose hits the geometry but bleeds 4× the deflection under load and risks elbow-singularity slowdown. If your measured tool position deviates more than 0.2 mm from the predicted value, suspect three things in order: (1) link-length calibration drift — the as-built a2 or a3 differs from the controller's stored value by more than 0.05 mm, common after a collision; (2) base-frame misalignment — the robot mounting plate has shifted relative to the work fixture, usually visible as a constant offset in one axis; (3) thermal expansion of the upper arm casting on a freshly-started cell, which can shift reach by 0.1-0.3 mm in the first hour before the arm reaches steady-state temperature.
When to Use a PUMA Arm and When Not To
The PUMA architecture isn't always the right choice. SCARA arms and delta robots beat it on speed and stiffness for tasks where you don't need full 6-DOF orientation. Cartesian gantries beat it on absolute accuracy and large flat workspaces. Here's where each one wins.
| Property | PUMA Arm (6-DOF Articulated) | SCARA | Delta Robot |
|---|---|---|---|
| Degrees of Freedom | 6 (full position + orientation) | 4 (X, Y, Z, yaw only) | 3-4 (position only, optional yaw) |
| Repeatability | ±0.05 to ±0.1 mm | ±0.01 to ±0.02 mm | ±0.05 mm |
| Typical Cycle Time (pick-and-place) | 1.5-2.5 s | 0.4-0.8 s | 0.25-0.5 s |
| Payload Range | 3-200 kg | 1-20 kg | 0.1-8 kg |
| Workspace Shape | Spherical, can reach around obstacles | Cylindrical, top-down only | Inverted dome, top-down only |
| Stiffness at Full Reach | Low (serial chain compliance) | Moderate | High (parallel kinematics) |
| Cost (industrial-grade, 5-10 kg payload) | $30,000-$60,000 | $15,000-$30,000 | $25,000-$45,000 |
| Best Application Fit | Welding, assembly, painting, complex orientation tasks | High-speed flat assembly, screw driving | Ultra-fast pick-and-place from conveyors |
Frequently Asked Questions About PUMA Arm
You're hitting the manipulability ellipsoid collapse near the elbow singularity. As θ3 approaches 0° (arm nearly straight), the Jacobian determinant drops toward zero, which means small Cartesian moves require huge joint velocities. The controller detects this and throttles back to keep joint speeds within limits.
Fix it by reposturing — bend the elbow more by lowering the base mount or moving the work closer. Most PUMA controllers expose a manipulability metric in the path planner; aim to keep it above 0.3 of nominal across the whole trajectory.
Three usual suspects. First, check whether the arm has taken a hit — even a soft collision can shift link offsets by 0.1-0.2 mm without triggering the collision-detect threshold. Second, look at gearbox temperature; harmonic drives expand thermally, and a wrist that was calibrated cold reads differently at operating temperature.
Third, and most overlooked, is the work fixture itself. If the fixture is bolted to the same baseplate as the robot and someone re-shimmed it, your robot-to-work transformation is now wrong even though the robot's internal kinematics are fine. Re-run a 4-point fixture frame teach before assuming the robot drifted.
SCARA, almost always. A SCARA's vertical Z-axis is a prismatic joint with stiffness an order of magnitude higher than a PUMA wrist resolving the same vertical motion through joint angles. For top-down insertion you don't need the orientation flexibility a PUMA gives you.
The exception is when the insertion axis isn't purely vertical — if the part needs to go in at 15° off vertical, or you have multiple insertion angles in the same cell, a PUMA's full 6-DOF wrist earns its keep. Below that, SCARA wins on cost, speed, and stiffness simultaneously.
No — that's wrist singularity escape. When joints 4 and 6 align (joint 5 near 0°), the two axes become collinear and the controller can't distinguish their contributions to orientation. To leave the singularity it has to make an arbitrary choice, and many controllers default to spinning joint 4 or joint 6 through 180° at full speed.
The fix is path planning — never command a Cartesian path that crosses joint 5 = 0° unless you've explicitly programmed a joint-space move through that region. Modern controllers (FANUC, ABB, KUKA) have a 'singularity avoidance' mode that detours around the singular configuration, but it slows the path and can fail if the start and end poses are on opposite sides of the singular surface.
The 7th axis buys you redundancy — for any tool pose, there's a 1-parameter family of joint configurations that achieve it, so you can reposture mid-path to avoid obstacles, singularities, or joint limits. That's worth the extra cost when you're working in cluttered spaces (medical, collaborative cells, around fixturing) or need to keep the elbow out of a specific zone.
For an open workcell with clear approach paths and no humans nearby, the 7th axis adds cost, mass, and one more joint to calibrate without buying you anything practical. Stick with 6-axis PUMA topology unless you can name a specific obstacle-avoidance or human-collaboration reason for the extra DOF.
Repeatability and path accuracy are different specs. Repeatability is how closely the arm returns to the same taught point — single-pose, low-speed. Path accuracy on a continuous trajectory is dominated by joint coordination errors, servo lag, and dynamic deflection, none of which the repeatability number captures.
For circular paths specifically, watch for joint reversals — if the path crosses a configuration where joint 4 or joint 6 changes direction, harmonic-drive backlash and motor-stiction transitions show up as a small bump or flat spot on the circle. Slow the path down by 50% as a diagnostic; if the error shrinks proportionally, it's servo dynamics, and if it stays the same, it's mechanical backlash.
References & Further Reading
- Wikipedia contributors. Programmable Universal Machine for Assembly. Wikipedia
Building or designing a mechanism like this?
Explore the precision-engineered motion control hardware used by mechanical engineers, makers, and product designers.
