Knowledge · Robotics

What determines 3D bin-picking cycle time?

The relevant measure is not the speed of one scan or robot move, but the production time required for one correctly picked and accepted part. Failed attempts, new captures and recovery therefore belong in the same time budget.

Robot cell used to measure picking cycle time

The right metric

Measure time per correct pick

First choose clear start and end points. A cycle might begin when a new capture is released and end when the part is correctly placed at its destination and the next cycle may start.

Measure a representative production window and divide total elapsed time by the number of accepted picks. Include waiting, missed picks, repeated captures and automatic recovery. State explicitly which planned breaks or external line stops were excluded.

Measurement model

Time per correct pick = observed production time / accepted picks

Use the same definition for every product variant and bin condition. Report attempts and accepted picks separately. If no pick is accepted, record the failed run and elapsed time rather than dividing by zero.

Time budget

Where the time actually goes

1. Observe

Trigger, illumination or projection, capture and transfer of depth data. Vibration or motion may require another capture.

2. Estimate the part pose

Process points, separate parts and determine position and orientation. Product variation and limited visibility affect computation time and confidence.

3. Validate a grasp

Select a reachable grasp and verify that the gripper, part and approach remain clear. The technical detail belongs ingrasp planning.

4. Move the robot

Approach, pick, withdraw, transfer and place. Path length, safe speeds and robot configuration often account for a substantial part of the cycle.

5. Confirm the pick

Jaw position, vacuum pressure or an additional check confirms that one part was correctly picked and placed.

6. Recover

A failed attempt may require returning the part, measuring again, trying another grasp or requesting operator assistance. These exceptions determine effective output.

Calculating achievable capacity

Theoretical speed is not line output

Use the same time boundary in the cycle and capacity calculations. If measured elapsed time already includes waiting, failed picks and recovery, those losses are already reflected in time per accepted pick. Do not apply an availability factor for the same losses again. Compare the resulting output with supply and downstream constraints.

Calculation model

Observed accepted output per hour = 3,600 × accepted picks ÷ total elapsed seconds

Illustrative inputs: a 3,600-second window including recovery produces 450 accepted picks. Time per accepted pick is 8 seconds and observed output is 450 per hour. Multiplying by availability again would count included losses twice. A separate planning model may use operating-only seconds per accepted pick and availability = operating time ÷ scheduled time, provided each loss belongs to only one term. Report excluded breaks and external stops explicitly.

Business case

Compare achievable output with manual hours per shift, labour cost, intervention time, changeover, maintenance, investment and demonstrably avoided scrap or downtime. A payback model is useful only when these inputs come from the actual production process.

Targeted improvement

Optimise the critical path, not every step in isolation

Run tasks in parallel

The next capture can sometimes occur during placement or withdrawal. First verify motion in the image, safety and the actual critical path.

Reduce the search space

Use the relevant bin volume and prevalidated grasps, provided they still cover product variation and reachability.

Shorten robot travel

Position the bin, camera and destination to avoid long moves and awkward wrist poses without compromising access or safety.

Design recovery first

Define when the cell tries an alternative, captures again, changes the part presentation or requests help. A short nominal cycle cannot compensate for frequent long interruptions.

Representative trial

Build the capacity claim on real production data

Share part variants, bin dimensions, required output, shift pattern and adjacent process steps. A trial can then record time per correct pick, availability, recovery and constraints.