Eye-in-Hand or Fixed Camera? Compare the Required View

Mark the feature the camera must see and the instant at which the robot needs that information. A camera position is useful only in relation to that view. The common mistake is to choose between a fixed camera and a wrist-mounted camera before anyone has listed what the robot must actually see and when. Once the required views are on paper, the mounting question usually narrows on its own.

Build the view-availability matrix before choosing a mount

The matrix lists the task stages and, for each, the feature that must be seen. Keep only the stages this project actually needs to image. Filling a stage that does not apply just adds rows that never influence the decision.

Mount position determines how the view moves, not an automatic accuracy ranking. A wrist-mounted camera moves with the end effector; a fixed camera observes from a stationary mount. Compare effective field of view, working distance, resolution, optics, occlusion and the feature the task must resolve for each proposal. In Zivid's hand-eye calibration description, eye-in-hand relates the camera to the end effector, while eye-to-hand relates the fixed camera to the robot base. The relevant transforms and their maintenance owners must match the selected arrangement.

Stage and required information	Fixed camera: what to verify	Wrist camera: what to verify	Evidence and conclusion
Identification feature at arrival	Whether the whole arrival zone falls inside the effective view	Whether enough information exists to locate before the move to the inspection pose	Installation drawing / sample image; unprovided means pending
Locating feature before grasping	Whether the approaching tool blocks the needed feature	Whether the camera can reach an observation pose without disturbing the tool motion	Motion sequence drawing / maker limits
Side features of a variant (only if the task needs them)	Whether the same fixed view keeps the critical feature	Whether the candidate observation move is within the working space	Variant images and a test design
Restart after tool or camera maintenance	Which frame relation or mounting position must be rechecked	Whether pose and cabling need reconfirmation after remounting	Owner and recovery record

Every empty cell is a blank to fill with evidence, not a silent "visible." A cell with no supporting drawing reads as unverified.

Read the fixed and wrist-mounted options against the same task

The comparison only means something when both options are judged against the same matrix. Two thought experiments show how the matrix is read. Both are illustrations with stated assumptions; neither claims a site-verified result.

A repeatable view from a fixed position. Assume the required feature is on the top face and the cell keeps the part in a fixed orientation. A fixed camera that always sees that face meets the need without arm motion.
A view that requires camera movement. Assume the required feature is on the side and the approach pose hides it from any fixed position. Here the wrist-mounted option earns a look, but only if the arm can carry the camera to the observation pose without interfering with the tool.

A repeatable view from a fixed position

When the needed feature is stable in the world frame and nothing in the approach hides it, a fixed camera keeps the view simple and the moving mass low. The claim to verify is narrow: that the arrival zone stays inside the effective view across the pose variation the line allows.

A view that requires camera movement

When the feature is only visible from a moving vantage, the wrist option becomes a real candidate. It carries its own costs, covered below, and its permission comes from the arm being able to reach the observation pose with the tool attached. If that reach is not demonstrated, the option is a hypothesis, not a plan.

A required view neither proposal has demonstrated

A proposal with both a fixed and a wrist camera should be assessed as a separate architecture, not an automatic best of both. Identify which required view each camera supplies, how their coordinate relationships are established, and whether the task requires observations to be associated in time. If neither single-camera proposal demonstrates the hardest required view, record that gap before adding hardware.

Include the mechanical and maintenance cost of the view

A camera is not free to mount. A wrist-mounted unit adds mass and inertia at the end of the arm, its cable must survive the working motion, and its relation to the light source changes as the arm moves. These are integration questions, and the answers depend on the specific robot and tool.

Wrist load, cables and access

Check the added payload against the robot maker's allowed attachment limits, trace the cable route through the full motion, and confirm the camera has an unobstructed view at the observation pose. These are drawing-level checks, not field measurements, until the actual configuration is assembled.

Calibration ownership after a change

Recheck the affected transform when the camera mounting, its reference relationship or a relevant calibration datum changes. Ordinary commanded motion does not itself invalidate a rigid hand-eye transform. A tool change may instead require a TCP, offset or obstruction check; determine which relationship actually changed. Assign an owner and a recovery trigger for those checks. The wider practice remains with recalibration triggers and ownership.

Whether the task needs 2D or 3D imagery is a separate decision from where the camera sits, and choosing between 2D and 3D vision covers the sensor dimension while this article covers the mount. If reflective surfaces later turn out to be a visibility obstacle, lighting tests for difficult surfaces is the right place to test that condition rather than folding it into the mount choice.

Choose against the hardest required view

The decision falls out of the matrix rather than a preference. Read the rows into three exits:

Fixed view already demonstrated. If a fixed position is shown to see the required features at the required times, keep the fixed candidate without adding a moving camera merely for its own sake; compare the actual maintenance requirements of the proposals.
Movement required and wrist option permitted. If the feature demands a moving vantage and the wrist option satisfies the payload, reach, and maintenance conditions, keep the wrist candidate.
Required feature still unseen or ownership missing. If the necessary view is not demonstrated by either option, or no one owns the recalibration trigger, add evidence and do not force a choice.

"In view demonstrated" has to come from research or site material. This round carries no such measured material, so every "demonstrated" cell is a target to be confirmed, not a result.

Does one required view still lack evidence? Bring the feature image, arrival and approach poses, and mounting constraints to robot selection support so the candidate comparison starts with that unresolved view. Keep the camera arrangement, required observation pose and owner of the recovery check on the same record.