Lenses for Robotics: Choosing Machine Vision Optics by Robot Task and Mount Format

What lens should I use for each robot vision task?

Robot vision breaks into three optical problems: navigation, manipulation, and inspection-on-arm. Each has a different primary constraint, and the right lens follows from the task, not from the robot platform as a whole. A single robot often carries more than one camera, each running a different lens for a different job.

Navigation: wide field of view for situational awareness

Mobile robots and AMRs need to see a wide area around the platform to avoid obstacles and localize against a map. Navigation cameras typically run fisheye or ultra-wide M12 lenses in the 150-200 degree range, which reduces the number of cameras needed to cover the robot's perimeter. High distortion is an acceptable tradeoff here: obstacle avoidance does not require millimeter-precision position measurements, so a 150-200 degree fisheye field of view matters more than a tight distortion spec.

Manipulation and pick-and-place: working distance and depth of field

Robot arms picking parts from a bin or fixture need the lens matched to working distance and depth of field, not just field of view. Overhead bin-picking cameras commonly run 300mm-800mm working distance and need a field of view wide enough to see the full bin. Coordinate accuracy depends on distortion: a lens with several percent barrel distortion at the corner of the frame introduces a position error that can be the difference between a successful grasp and a missed one. Working distance is a design variable early in a project (brackets and camera mounts can still move), but it becomes effectively fixed once the mechanical design is locked.

Inspection-on-arm: resolution and low distortion at a fixed standoff

Cameras used for part inspection on a robot arm or at a fixed inspection station need enough resolution to resolve the defect size at the working distance, plus distortion low enough that measurements taken from image coordinates are trustworthy. Because these cameras are frequently at a fixed station rather than moving with the robot, C-mount lenses become practical: the weight penalty does not cost robot payload, and the adjustable iris helps hold focus across parts of varying height.

The table is a starting point; real robots mix roles. A mobile manipulator might carry a fisheye lens for navigation and a low-distortion lens on the arm for grasping, running two different optical specs on the same platform. See the lenses for embedded vision guide for board-level integration details common to all three roles.

End-of-arm inspection on robot arms

Cameras mounted directly on end-of-arm tooling typically operate at 100mm-500mm working distance and move continuously with the arm. Weight is the primary constraint, vibration exposure is high, and distortion needs to stay low because there is no opportunity to calibrate out several pixels of barrel distortion on every cycle. The CIL034 is a reasonable default for this role: IP67-sealed metal barrel, essentially no measured distortion, and only 5.7g. Its hybrid glass/plastic construction is acceptable for this role given the sealed metal barrel and its low measured distortion; for deployments that need all-glass construction as well, see the CIL052. Use a longer focal length, such as the CIL085 at 8.2mm, when the task needs more standoff distance or finer detail for defect detection.

Mobile robots and AMRs

Autonomous mobile robots need obstacle detection and situational awareness across a wide area. Outdoor AMRs generally need IP67 sealing since dust and rain are real; indoor warehouse robots do not strictly need sealing, but it is inexpensive insurance against spills and dust. The CIL232 covers a genuine 181-degree field of view at 12g and IP67, adding almost nothing to mobile platform payload. The lighter, unsealed CIL391 is a reasonable option for indoor AMRs that do not need sealing.

Collaborative robots (cobots)

Cobot cameras are typically wrist-mounted or positioned above a shared workspace, with moderate working distances around 200mm-600mm. Hand-eye calibration accuracy depends directly on distortion, so use low-distortion lenses and re-run calibration whenever the lens is adjusted or replaced. Higher distortion measurably degrades calibration residuals; very low distortion gives diminishing returns from software correction. The CIL052 at -0.1% distortion contributes negligible distortion-related calibration error.

Gantry and fixed-station inspection

In gantry or conveyor setups, the camera is fixed and the part moves through the field of view, so the camera does not draw against robot payload. Larger C-mount lenses become practical, higher magnification is available for defect detection, and vibration comes from the conveyor and factory floor rather than from camera motion, generally lower amplitude and more predictable than end-of-arm vibration. The CIL556 50mm C-mount lens, 17-degree field of view at its 19.4mm image circle, is a reasonable choice for high-magnification defect inspection at a fixed station, with -0.02% distortion that is effectively negligible.

Why the lens matters for robot vision accuracy

Sensor and compute platform selection get most of the design attention on a robot vision project. The lens is often an afterthought until it causes a field problem, most commonly a coordinate mapping error, a calibration drift, or an obstacle-detection blind spot.

Resolution at working distance, not just on the bench

MTF bench data describes peak optical performance under ideal conditions. What matters on a robot arm is resolution at the actual working distance, across the full field the task needs. A lens with high center MTF and weak edge performance gives accurate results near the image center and degraded results at the edges, which is often exactly where bin-picking coordinates or part edges live.

Distortion and coordinate accuracy

Lens distortion introduces a position error that grows from the image center toward the edge. For pick-and-place, several percent of barrel distortion at the corner of the frame can translate to multiple millimeters of positional error at the gripper. Low-distortion lenses with distortion under roughly 0.5% remove most of this error without requiring a per-pixel software correction on every frame.

Vibration and mechanical stability

Vibration and mechanical shock are a real consideration for any camera that moves with a robot. Glass elements in a metal barrel are generally more dimensionally stable than plastic elements, so all-glass construction is one factor worth weighing for a lens that will live on a moving arm or chassis. Check the vibration and shock ratings on the datasheet against your duty cycle.

Weight on end-of-arm tooling

Every gram on an end-of-arm tool counts against the robot's rated payload. A typical C-mount lens weighs 50g-200g; a typical M12 lens weighs 3g-15g. On a robot arm with a modest payload rating, the lens alone can represent a meaningful share of the payload budget once the gripper, cabling, and any onboard illumination are accounted for. Use M12 lenses for any camera that moves with the robot.

Environmental sealing

Machine shop and food processing environments expose lenses to coolant mist, wash-down water, and particulate. IP67-rated lenses are dust-tight and rated for temporary immersion. For robots operating in these environments, sealing is worth specifying; for a controlled indoor cell, it typically is not required. Not every M12 lens is sealed, so verify the datasheet rather than assuming.

Datasheet gaps: MOD and working distance are not always published

Minimum object distance and rated working distance are not published on every lens datasheet, and manufacturers are not consistent about which convention they use when they do publish it. Commonlands measures MOD from the front of the lens to the object, following Edmund Optics convention. For C-mount, MOD is a hard limit set by the end of cam travel: the focus ring physically stops. For M12, there is no equivalent hard stop; within its thread travel, the rigid assembly can usually be threaded outward to focus the center of the field at short working distances, so the practical limit is typically field curvature and astigmatism at the corners rather than a mechanical stop. If a datasheet does not list MOD, ask before assuming a lens will focus at your intended working distance.

Lenses for stereo vision and depth sensing

A stereo camera uses two lenses separated by a known baseline to capture the same scene from slightly different viewpoints, then estimates depth from disparity, the pixel offset between where a given point appears in the left and right image. The relationship is Z = (f × B) / d, where Z is scene depth, f is focal length in pixels, B is baseline, and d is disparity. Because disparity appears in the denominator and shrinks with distance, precision degrades with distance: a fixed disparity error produces a depth error that grows with the square of distance, so a 1-pixel disparity error at 500mm produces a much smaller absolute depth error than the same error at 2000mm.

Longer focal length increases disparity sensitivity, improving depth precision, but narrows field of view and shrinks the overlap volume shared by both cameras. A wider baseline improves depth discrimination but pushes out the nearest distance at which both cameras can see the same object. Choose baseline and focal length together against the working volume you actually need to cover, and verify shared coverage with the field of view calculator before committing to hardware.

Why matched lens pairs matter for disparity accuracy

Stereo depth assumes both lenses behave consistently: the same scene geometry should produce proportionally equivalent image geometry in both cameras. If the left lens has a slightly longer effective focal length than the right, objects at a given depth produce different magnification in each image, and stereo calibration can only partially absorb the residual. Two lenses marketed under the same SKU can still differ measurably in effective focal length and distortion due to manufacturing tolerances, small for a single-camera application but potentially significant in a precision stereo system. For demanding stereo accuracy, characterize both lenses individually rather than assuming the catalog number guarantees optical equivalence.

Distortion mismatch between the two lenses degrades stereo rectification, the preprocessing step that aligns both images so corresponding points fall on the same horizontal scan line. When distortion profiles differ, rectification residuals remain, forcing the stereo matching algorithm to search a wider vertical range and raising the chance of false matches near the field edges. Asymmetric field sharpness has a similar effect: if one lens has weaker corner MTF than the other, the matching algorithm compares a sharp patch against a blurred one, which lowers match confidence in that region.

Synchronization: a system-level constraint, not a lens spec

Stereo depth also depends on both cameras capturing the same instant. Hardware-synchronized trigger lines, or a global-shutter sensor pair with a shared clock, keep left and right frames aligned in time; without synchronization, a moving scene introduces apparent disparity that has nothing to do with depth. This is a sensor and camera-system decision the lens does not control, but it belongs on the same checklist as lens matching, because a perfectly matched lens pair cannot fix depth error introduced by unsynchronized frames.

M12 vs C-mount for stereo builds

M12 is the practical choice for compact embedded stereo modules: two lenses fit on a small sensor pair without adding meaningful mass, and current-generation M12 lenses cover sensors up to roughly 1/1.7" with distortion from near-zero to roughly 2%, acceptable for calibrated stereo. C-mount becomes the better choice once the sensor exceeds roughly 1/1.7", the assembly needs a rigid, precisely located optical path, or depth-of-field control across a variable working range matters. The fixed 17.526mm C-mount flange gives a repeatable mechanical reference that simplifies stereo rig alignment; the tradeoff is size, weight, and cost.

Stereo prototype checklist before locking the design

1. Verify baseline and focal length at maximum working distance. Use Z = (f × B) / d with your minimum detectable disparity to compute depth error at the far end of the working range.
2. Confirm shared field of view at minimum working distance. Both cameras must see the full scene width at the nearest required depth; check with the FoV calculator.
3. Check image circle against sensor diagonal on both cameras. Use actual image circle spec, not the rated sensor format label.
4. Verify distortion is specified at the full sensor image circle, not a smaller reference diameter.
5. Check corner MTF on both lenses, not just center resolution. Asymmetric corner sharpness between left and right degrades matching at the periphery.
6. Lock lens focus mechanically after calibration. Rotating an M12 lens after stereo calibration shifts the focal plane and invalidates the calibrated distortion model.

For the full stereo depth math, rectification residual analysis, and CRA-versus-microlens interaction, see the chief ray angle and mismatch guide and how to read MTF curves.