Robotics Fundamentals
Reading a Robot's Specifications
What payload, reach, repeatability, accuracy, resolution and cycle time actually mean on a robot datasheet. Plus why repeatability and accuracy are not the same number.
beginner · 12 min read
Lesson content
A robot datasheet is a short list of numbers and almost all of them are misread the first time. This lesson walks the list, because choosing a robot badly is nearly always a matter of trusting the wrong number on it.
We start with the two that get confused most often.
Those two targets are the whole confusion, so it is worth saying the difference in one sentence each.
Repeatability asks: when the robot goes back to a point, how close is it to where it went last time?
Accuracy asks: is that point the one you actually asked for?
They are independent. A robot can be brilliant at one and hopeless at the other, which is exactly the left-hand target.
- Payload
The mass the robot can carry at the tool flange, including the gripper itself.
A 10 kg robot holding a 4 kg gripper has 6 kg left for the part. Half of all payload mistakes are forgetting the gripper.
- Reach
The greatest distance from the axis of the base to the tool centre point.
A single radius on one plane. The true workspace is a shell with dead zones, so read the work-envelope drawing rather than this number.
- Repeatability
How closely the robot returns to a point it has been taught, over and over.
This is the number industrial robots are actually sold on, typically ±0.02 to ±0.1 mm, because most robots are taught by pointing rather than by coordinates.
- Accuracy
How closely the robot reaches a point you gave it in coordinates it has never been taught.
Usually ten to a hundred times worse than repeatability and often not published at all. It matters the moment you generate paths from CAD instead of teaching them by hand.
- Resolution
The smallest movement the control system can command or measure, which is one encoder count at the joint.
Sets the floor for everything else. You cannot be repeatable to finer than your resolution.
- Cycle time
How long one standard move-and-return takes, usually quoted for a 25 mm–300 mm–25 mm path with a 1 kg load.
Only comparable between robots when the test path and the load match. Vendors choose flattering paths.
Darts, not dartboards
Throw six darts at a board.
All six land in a tight cluster in the bottom-left corner. You are repeatable but not accurate and honestly that is fine. Shift your aim once and the whole cluster lands on the bullseye.
Now the other case. The six darts scatter all over the board, but their average is dead centre. You are accurate but not repeatable. There is no aim correction that fixes this. Every throw is a surprise.
That is why industry pays real money for repeatability and mostly shrugs at accuracy.
Why is accuracy so much worse than repeatability?
Because they ask the robot to do two completely different things.
Repeatability only asks it to go back to a state it has already been in: the same encoder counts it visited last time. No maths involved.
Accuracy asks it to turn coordinates into joint angles. That means the controller's internal model of link lengths, joint offsets and gearbox flex has to match the actual steel sitting on the floor.
So manufacturing tolerance, thermal expansion and the arm sagging under its own weight all land squarely in the accuracy number. None of them touch repeatability at all, because they repeat identically every cycle.
| Class | Payload | Reach | Repeatability | Typical use |
|---|---|---|---|---|
| Delta picker | 1–3 kg | 1100 mm | ±0.1 mm | Food packing at 150+ picks/min |
| SCARA | 3–20 kg | 400–1000 mm | ±0.01–0.02 mm | PCB and small-part assembly |
| 6-axis industrial | 5–300 kg | 700–3100 mm | ±0.03–0.2 mm | Welding, palletising, machine tending |
| Collaborative arm | 3–16 kg | 500–1300 mm | ±0.03–0.1 mm | Shared workspace, low-volume tasks |
Repeatability gets worse as reach and payload grow, because a longer lever amplifies every joint error.
The math
Payload is not one number. It is a moment. A motor at the shoulder has to hold up the mass times the distance it hangs out. Double the distance and you have doubled the load on that motor even though the mass has not changed.
τ = m × g × L
- τ
- torque the joint must produce, in newton-metres (N·m)
- m
- mass at the end, in kilograms: part plus gripper
- g
- 9.81 m/s², gravity
- L
- horizontal distance from the joint to that mass, in metres
With real numbers
A 5 kg part held 1.2 m out from the shoulder:
τ = 5 × 9.81 × 1.2 ≈ 59 N·m.
Pull the same part in to 0.6 m and the shoulder only sees ≈ 29 N·m, half the strain, for free. This is why load-versus-reach charts curve downward and why bringing the work closer to the base is the cheapest performance upgrade there is.
Static payload is not the whole story
The number on the datasheet is what the robot can hold. What it can move quickly is always less.
The reason is simple. Getting a load moving adds inertial torque on top of the gravity torque that was already there.
So when a part is heavy and hangs off to one side, the vendor will ask for three numbers rather than one: its mass, how far its centre of gravity sits from the flange and its moment of inertia. Those three together decide whether your cycle time is possible.
Two more numbers appear on nearly every datasheet and are worth ten seconds each.
Duty cycle is how much of the time the robot is allowed to run flat out before it has to cool down.
Protection rating is IP54, IP67 and so on. This decides where the robot is allowed to live: a paint booth, a food line that gets hosed down, or only a nice dry cell.
Sizing payload from the part alone.
Payload = part + gripper + hoses + camera + anything else bolted past the flange. Then leave 20 % headroom.
Comparing cycle times between vendors.
Only meaningful if the test path and test load are identical. Ask for a simulation of your path instead.
Assuming ±0.02 mm repeatability means parts land within 0.02 mm of where you want them.
It means they land within 0.02 mm of each other. Where that cluster sits relative to your CAD model is accuracy, which is a different and larger number.
Key takeaway
- Repeatability is the same place every time. Accuracy is the right place. Repeatability is nearly always the better number.
- Payload includes the gripper and falls as the arm reaches further, because torque is mass times distance.
- Reach is one radius rather than a sphere, so read the work-envelope diagram instead.
- Resolution is the floor: nothing downstream can be finer than one encoder count.
- Cycle times are only comparable when the path and load are identical.
Finished this lesson?
Guest progress is saved on this device.