Where is the fly?
Wide channel view → pickup coordinates
CLEMSON · SENIOR-DESIGN PROTOTYPE · 2026
Robotics for Drosophilaresearch.
Automated fly handling. Quantitative behavioral assays. Computer vision, custom electronics, and robotics built to reduce repetitive work in genetics research.
1st placePeople’s Choice · Clemson Senior Design ShowcaseAVI PATEL + TEAM 8ROBOTICS / VISION / EMBEDDED SYSTEMS
THE COMPLETE PLATFORMExplore full 3D ↗
Loading interactive 3D…
One platform, from fly handling to behavioral data.
Photo-informed reconstructionInspect assemblies + watch the sequence ↗>97%YOLO sexing accuracy 45 / 46 test images
3Camera perspectivesOne coordinated platform
7Cross-functional buildersClemson senior design
✳1st · People’s ChoiceClemson · 2026 showcase
01TOOLS → WORKING HARDWARE
Four disciplines.
One research platform.
Hover over a discipline to see what we built with it. Every tool connects to a real project artifact.
Model, fabricate, test, repeat.
Modular structures, replaceable channels, and interchangeable vacuum nozzles turn CAD revisions into testable hardware. The interactive Blender reconstruction makes the finished assembly explorable here.
See the design record02WHY AUTOMATE THIS WORKFLOW?
Less repetitive handling.
More time for the experiment.
Genetic studies depend on preparing the right fly groups and measuring how they behave. Both require repeated, precise work.
Our prototype connects sorting, controlled imaging, and climbing-video analysis. It targets the work between a research question and usable experimental data.
RESEARCH TASK
Start with a mixed vial of 40–50 flies.
WHAT THE PLATFORM CONTRIBUTESOrganize experimental groups with camera-guided handling and physical routing.
Explore the report~2flies / minute
Reported prototype sorting rate
Reported prototype sorting rate
A working handling prototype, paired with quantitative video analysis—not just a model prediction. Presentation results ↗
The goal is faster experimental preparation and analysis. The sorting rate is a prototype result, not a measured speedup over a manual workflow.
03FOLLOW ONE FLY
How the prototype
handles and measures flies.
Start with a fly. End with a result a researcher can inspect. Follow the physical path—and the data it creates.
01 / PREPARATION + HANDLING
Give tiny specimens a predictable starting point.
Sedated flies are loaded manually into a printed pickup channel. Channel geometry and CO₂ distribution hardware were iterated to support consistent presentation for camera-guided handling.
AT THIS STATIONPrinted channel + CO₂ manifold
~25flies · channel design capacity
Report · p. 6 ↗
01 / LOAD
Loading channel + CO₂ hardware
LOOK CLOSER / 01
Physical CO₂ manifold and channel assembly
THE RESEARCH CONNECTION
Consistent presentation makes the next image—and the next pickup—more predictable.
INPUT
→Manually loaded, sedated fruit flies
ENGINEERING
→Channel geometry + CO₂-supported handling
OUTPUT
Specimens presented within the pickup channel
The walkthrough connects separate subsystem demonstrations, not a continuous run of the same fly. At the assay, individual routing gives way to measuring a prepared vial group.
THREE CAMERAS. THREE DIFFERENT QUESTIONS.
A wide view to locate. A close-up to classify. A fixed view to measure.
Which sex is it?
Controlled close-up → classification
How does it move?
Backlit vial view → behavioral analysis
PROTOTYPE TEST FOOTAGE
Handling tests and tracking.
Three short clips, from physical handling to behavioral tracking. Play, scrub, or open full screen.
01 / COORDINATED MOTION19 SEC
Gantry motion
The belt-driven gantry carries the vacuum head across the channel.
Open clip ↗02 / VACUUM + GEOMETRY15 SEC
Vacuum pickup
A closer look at the nozzle and the physical handling interface.
Open clip ↗03 / OFFLINE VISION10 SEC
Climbing-assay tracking
Recorded assay footage with actual tracking boxes, identifiers, and thresholds.
Open clip ↗04MEASURED, NOT ASSUMED
What we tested
and what remains.
Three evaluations, three distinct scopes.
Select a result to inspect what was actually tested.
YOLO SEX CLASSIFICATION
>97%
45 of 46 test images classified correctly.
A focused diagnostic on a test sample: all 19 female images and 26 of 27 male images were classified correctly.
CLASSIFIED CORRECTLY45 / 46
Correct classification Misclassified
19 / 19 female · 26 / 27 male
Aggregate tally, not test order.
YOLO image diagnostic · separate from the hybrid evaluationREPORT P. 14 ↗
05ENGINEERING DETAILS
From a drawing
to a working prototype.
Explore the CAD, control board, vision pipeline, and assay software.
CAD → FABRICATION
Designed to be built. Designed to be changed.
A modular, three-section printed table brings the channel, imaging station, and assay into alignment. Replaceable parts make physical iterations practical.
- ↳Modular 3D-printed structure
- ↳Belt-driven single-axis motion
- ↳CO₂ manifold + vacuum pickup
DESIGN ITERATIONS
What changed during testing.
The hard part was connecting reliable perception to imperfect physical hardware.
Channel geometry and CO₂ distribution
What changed +
BOTTLENECK
Flies could cluster, move, or become difficult to isolate.
ITERATIONIterated channel geometry and CO₂ distribution alongside the pickup interface.
Design record · p. 6 ↗Nozzle length and transfer reliability
What changed +
BOTTLENECK
Pickup, release, and lip collisions created different failure modes.
ITERATIONTested nozzle lengths and interfaces; the final nozzle was 19.5 mm long.
Design record · p. 8 ↗Lighting, focus, and classification
What changed +
BOTTLENECK
Focus, lighting, and background changes affected the image distribution.
ITERATIONRefined the imaging environment and combined learned appearance with segmented morphology.
Design record · p. 15 ↗Recording, offline analysis, and vibration
What changed +
BOTTLENECK
Frame-by-frame processing competed with acquisition; vibration also coupled through the table.
ITERATIONSeparated recording from offline analysis and investigated the vibration-holder design.
Design record · p. 12 ↗THE STACK
Python / OpenCV / YOLO / Random Forest / Raspberry Pi 5 / KiCad / 3D printing
06PROJECT RESOURCES
Report, video,
and presentation.
Original materials from the team’s senior-design project.
Download the editable presentation (PPTX) ↓
CLEMSON SENIOR DESIGN II / SPRING 2026CLEMSON TEAM 8A seven-person
A seven-person
senior-design team.
AI-Powered Robotic Platform for Genetic Screening:
Built an autonomous system that identifies, sorts, and evaluates fruit flies for genetics research. Integrated computer vision, embedded controls, and custom hardware to achieve ~97% classification accuracy, earning Clemson’s People’s Choice Award with a cross-functional group of 7 people.
Seven-person senior-design prototype · YOLO diagnostic: 45/46 test images
✳
People’s Choice AwardClemson Senior Design II Showcase
Clemson University · Institute for Human Genetics sponsor