How a vision robot trims hemp flower

Harvested flower rides a conveyor under a camera that photographs each bud and finds the leaf that should come off. A controller turns those pixels into cut points, and a robot arm with a small scissor head snips the leaf away while the trimmed bud rides on to a product bin and the clippings fall into a separate trim stream. The cell only works if both the camera and the arm have enough time per bud.

Controls

Labels

Animation speed

Conveyor speed

Camera frame rate (processed)

Frames the vision model actually analyzes each second, not the sensor's raw rate.

Cuts per bud

Step through it

Flower on the beltHarvested buds are spaced out on a conveyor, about 15 cm apart in this model. Belt speed sets how many buds arrive each hour.
The camera looksA camera under a ring light photographs each bud as it passes. The vision model separates leaf from flower and picks cut points. The view cone turns red when there are too few frames per bud.
The arm cutsThe robot arm moves its scissor head to each cut point. More cuts per bud remove more leaf but stretch the cycle time for every bud.
Two streamsTrimmed buds ride on to the product bin. Clippings drop down a chute into a separate trim bin, which processors weigh and track as its own material stream.
Push the beltSpeed the belt to 12 m/min and drop the camera to 5 frames per second. The robot falls behind and the camera sees each bud only a handful of times.

Two clocks that both have to keep up

Buds arrive at belt speed divided by spacing. With 15 cm between buds, a 3 m/min belt brings 1,200 buds an hour. The arm needs time to move to each bud plus time for each snip; this model assumes 0.6 s to reposition and 0.3 s per cut, so 7 cuts take 2.7 s and the arm can finish about 1,330 buds an hour. Throughput is the smaller of the two numbers.

The camera has its own budget. A bud stays in a 20 cm view window for that distance divided by belt speed, so frames per bud equals processed frame rate times that time. This model asks for at least 10 frames so the software can see each bud from several angles as it shifts on the belt. Both the timing values and the threshold are teaching assumptions, not specifications of any machine.

Try it: set cuts per bud to 16 and watch the bottleneck switch to the robot, then lower the belt until the spare capacity comes back.

The trim estimate is simplified

Each extra cut removes a share of the leaf that is left, so this model uses leaf removed = 1 minus e to the power of (minus cuts divided by 6). It assumes leaf makes up about 20 percent of wet bud weight. Real ratios vary widely by cultivar, how the plant was grown, and whether the flower is trimmed wet or dry.

Licensed processors track the trim stream separately from finished flower, so a cell like this weighs both bins rather than estimating.

Where Avid uses this

Avid works on agricultural robotics, where the same question of camera frames and actuator time per plant decides whether a vision system can run in real time.

Sources

Post-harvest processing of hemp flowersOregon State University Extension Service
Edge-AI-Based Vision-Guided Robotic System for Selective Trimming of Unhealthy Leaf Clusters in Controlled Agricultural EnvironmentsMichigan Technological University Digital Commons (IEEE AEECT 2026 paper)

How to cite this model

Avid Solutions International. (2026, November 11). How a vision robot trims hemp flower [Interactive 3D model]. Avid Field Lab. https://avidsolutionsintl.com/explore/robotic-flower-trimmer

Last updated November 11, 2026

Keep exploring

Hemp & CannabisHow cold ethanol washes oil from hemp Hemp & CannabisHow heat activates cannabinoids in a reactor Hemp & CannabisHow hemp flower is dried after harvest Hemp & CannabisHow a lab tests hemp for THC