Autofocus vs fixed focus lenses on mobile robots: vibration and blur

Mount a camera on a mobile robot — an AGV, AMR or research rover — and the first questions are power and mounting; the second is focus. Consumer-style autofocus modules are built for phones and webcams that sit still, while a robot accelerates, brakes and crosses floor seams that shake the whole camera. The numbers below decide autofocus versus a fixed, locked lens for your platform: the smear formula, the exposure cap and a depth-of-field budget, worked through for a 2/3" 5 MP camera at 0.5 m/s. You will leave with a decision rule for each case and a short verification plan.

Will an autofocus lens survive the vibration on a mobile robot?

No: a compact autofocus module typically needs 100-500 ms to settle one focus correction, and a robot moving at 0.5 m/s covers 50-250 mm in that window, so every focus it lands on is out of date. The module is tuned for a handheld camera that sits still; a chassis that accelerates, brakes and bounces breaks both assumptions at once.

Two things go wrong together. The contrast the loop measures changes with every bump, so it never concludes the frame is sharp and keeps stepping the motor. Even a correct step arrives tens of milliseconds late, by which time the object distance has moved again. The result is hunting: a focus that oscillates around the right value and is wrong in every frame.

The fix is not a smarter algorithm. When the object distance stays constant to within half the depth of field, a fixed-focus machine vision lens set once and locked needs no focusing system at all; the sections below turn that into numbers. For capturing and locking a focus properly even at a static station, see locking the focus after autofocus.

What actually blurs the image on a moving robot?

Three mechanisms, ranked by how much they cost at typical rover speeds: drive motion during exposure, suspension vibration, and focus hunting — and at 0.5 m/s with a 1 ms exposure the first alone smears the image by about 11 pixels, which dwarfs the chassis shake.

Drive motion is plain geometry: during the exposure the robot translates, and the scene slides across the sensor in proportion to magnification. That is the same smear that softens edges on any moving part, and the conveyor math applies unchanged. Suspension vibration adds a smaller velocity term: a chassis bouncing 0.2 mm at 30 Hz peaks at 2π × 30 Hz × 0.2 mm ≈ 38 mm/s, an order of magnitude below drive speed, so it only matters once drive motion is handled.

Focus hunting defocuses rather than smears, and a stepping motor running during a long exposure can add a wobble of its own. A fourth mechanism is electronic: on a rolling shutter sensor the rows are read out over several milliseconds, so vibration shears and wobbles the frame even when the exposure itself is short.

How do you compute the blur budget before you build?

Take pixel smear as m·v·t_exp/p and cap the exposure at p/(m·v); for a 2/3" 5 MP camera with 3.45 µm pixels viewing a 120 mm wide field from a robot at 0.5 m/s, that cap is 94 µs, about a tenth of a common 1 ms default. Choose lighting to support that exposure, not the other way round; the exposure time versus shutter speed trade and the case for capping the auto exposure maximum cover that side.

Here m is magnification (sensor width divided by field width), v the robot speed, t_exp the exposure time and p the pixel pitch.

Worked example, with the arithmetic shown:

• Magnification: m = 8.8 mm / 120 mm = 0.0733 (a 2/3" sensor is 8.8 mm wide).
• Smear at 1 ms: the robot travels 500 mm/s × 0.001 s = 0.5 mm; on the sensor that is 0.0733 × 0.5 mm = 36.7 µm, and 36.7 µm / 3.45 µm = 10.6, about 11 pixels of smear.
• Exposure cap for a 1 px budget: t_exp = 3.45 µm / (0.0733 × 500 mm/s) = 3.45 / 36,700 µm/s ≈ 94 µs.
• Vibration share: 38 mm/s of chassis bounce during 94 µs moves the image 38,000 µm/s × 0.000094 s × 0.0733 = 0.26 µm, or 0.08 px. Negligible once the cap holds.

The decision follows directly. At 0.5 m/s you need roughly a 94 µs exposure, which takes strobed or strongly boosted light; at 0.1 m/s the same budget relaxes to 470 µs, which ordinary continuous lighting reaches. If you want the real chassis amplitude instead of an estimate, frame-to-frame tracking measures it from the video itself.

How much defocus can a fixed, locked lens tolerate?

Depth of field is the entire focusing budget: for the same 2/3" geometry at F5.6 with a two-pixel circle of confusion, total depth of field is about 15 mm, so any target that stays within roughly ±8 mm of the set distance needs no focusing system at all. Autofocus earns its place only when the object distance swings further than that.

N is the f-number, c the circle of confusion, m the magnification. With m = 0.0733 and c = 2 px = 6.9 µm:

• F2.8: 2 × 2.8 × 0.0069 mm × 1.0733 / 0.005373 = 7.7 mm total.
• F5.6: 15.4 mm total, that is ±7.7 mm about the set distance.
• F8: 22 mm total.

Stopping down past F8 starts to cost resolution on 3.45 µm pixels to diffraction, which the diffraction and aperture article quantifies, so the aperture is a trade, not a free dial. Measure the real distance swing across a mission — floor seams, ride-height change under payload, target height tolerance — and hold it inside half the depth of field. The calculator above runs the same numbers for your own sensor and lens.

Where does autofocus still earn its place?

Keep autofocus where the camera is still during capture: bench and kiosk stations, docking and charging points, and robots that stop, settle, focus and then drive with the lens locked.

That pattern suits platforms whose object distance changes from stop to stop beyond the depth of field, such as a service robot reading labels at different shelf depths. Slow rovers on genuinely smooth floors sometimes tolerate continuous AF, but expect hunting on low-contrast floors and treat it as best effort. Where each frame is captured sharp but the distance changes between frames, a liquid lens changes focus in milliseconds with no traveling focus group, which is far friendlier to a chassis than a stepper. For measurement stations, fixed-magnification optics remove focusing from the conversation entirely; the entocentric versus telecentric comparison covers that split.

The LS-TCL-0574 is a fixed-magnification, C-mount lens holding 0.4x at a 227 mm working distance on a 1.2" sensor.

Which lenses hold focus mechanically on a moving platform?

Choose lenses built to stay put: an anti-vibration series with a damped, locked focus cell, or any fixed lens whose focus ring you clamp or fix after setting. What matters on a chassis is that nothing in the barrel can migrate: no free-running motor, no loose ring, no gravity-loaded element.

Set the focus once at the working distance on a rigid bench, lock it, and mark the ring. Anti-vibration designs add internal damping so road shock cannot reseat elements; they are the default pick for anything that drives. Sensor format and resolution still decide the optical match, so filter the lens finder by those two before comparing series. Focus also drifts with temperature over a day's work, and the focus drift article shows how to tell drift from blur and when a re-set is due.

The LS-FFL-0475 is a 36.61 mm, 2/3", C-mount anti-vibration lens.

For cameras that stand further back, the LS-FFL-0914 is a 0.3x, C-mount lens with a 168 mm working distance, also for the 2/3" format.

How do you verify blur and focus before deployment?

Run two checks: a bench check on a rigid stand to set focus and confirm image quality, then a motion check at mission speed where the edge spread of captured frames tells you the real blur in pixels. Hold a 1 px budget for measurement tasks and accept 2-3 px for detection.

The bench check needs a mount that cannot flex. A professional slide-table stand such as the LS-MNT-0040 carries a 700 mm column and a 400 x 450 mm base for setting the exact working distance before anything moves.

Then verify on the chassis itself:

1. Lock the focus and mark the ring at the set working distance.
2. Set the exposure to the computed cap (94 µs in the worked example) and turn auto exposure off.
3. Drive at mission speed over a printed target and record frames.
4. Measure edge spread across a known straight edge and compare it with the pixel budget.
5. Repeat at temperature extremes and after a bump session; re-lock if the spread grew.

If the spread exceeds budget, slow the platform before adding light: speed enters the smear formula linearly, while exposure only sets how much light you must buy back. The camera stands and brackets range covers rigid mounting when the camera needs a bracket rather than a bench, and the systems topic page collects the related guides. To size the whole chain from field of view to pixel count, the lens configurator walks the same inputs in order. When the geometry fights back, send the numbers through the contact page and we will work them with you.