C-channel singulation

The question

How does Sorter V2 singulate LEGO pieces without vibration?

Traditional part feeders use vibration (bowl feeders, V-channels) to separate tangled parts. Vibration works but is loud, hard to control, and assumes roughly uniform part geometry. LEGO pieces vary wildly in size and shape. The project needed a quieter, more controllable alternative.

The mental model

A C-channel stage has two parts: a faceted rotor (not round) with an angled inner cone, and a static outer stator wall. Together they form a C-shaped channel that pieces sit in. When the rotor turns, pieces are pushed outward along the channel toward an exit guide.

The key insight: the drops between stages are the primary separation mechanism, not the rotation itself. Each inter-stage drop has a probability of untangling clumped pieces. Stack three stages and the probability of a clump surviving all three drops converges toward zero.

"You can actually take the entire v-channel concept and make it circular." — Spencer

"We found a way to do v-channels through circular motion without vibration. I have not found/seen anything like this online." — Marc

How it works

Component	Details
Stages	3 in series (2 likely sufficient; 3 is safety margin)
Cameras	3× OV9732 100° wide-angle, one centered above each channel
Calibration	ArUco 4×4 tags (3D-printable, two-color) define center, exit, and bulk boundary
Piece tracking	OpenCV MOG2 background subtraction — no training data needed
Anti-jam	5-level escalating algorithm: progressively stronger back-and-forth shaking
Color calibration	CIELAB comparison with reference card, UVC control adjustments
Lighting	50mm COB PCB LED in vertical post at center of each channel, side-mounted

Control loop

1. Camera detects pieces via MOG2 pixel-diff tracking.
2. Rotor advances until pieces reach the exit zone.
3. If a clump is detected (multiple contours too close), rotor reverses briefly.
4. Anti-jam escalation kicks in if reversal fails — five levels of increasingly aggressive shaking.
5. Single piece exits to the next stage or to the classification chamber.

Performance

Metric	Value
Current throughput	~330 pieces/hour (5.5 PPM over 42-minute sustained run)
MVP target	360 pieces/hour (6 PPM) — achieved March 2026
Ultimate target	1,000 pieces/hour (~17 PPM)
Theoretical ceiling	~20 PPM before carousel bottleneck
Bottleneck	Feeder physics, not software or classification
Singulation confidence (3 stages)	>99%

Trade-offs

• C-channels vs V-channels — V-channels use vibration for natural agitation but are loud. C-channels are quieter and more controllable via software, but require more physical length (3 stages) to achieve equivalent separation confidence.
• Circular vs linear — Circular motion gives a smaller footprint and simpler overhead camera geometry (no perspective skew). Linear paths would need longer travel distance.
• 2 vs 3 stages — Two stages likely suffice for most loads. The third stage is a safety margin that the team kept because the cost (one more rotor + camera) is low relative to the confidence gain.
• MOG2 vs YOLO for tracking — MOG2 background subtraction is faster, needs no training data, and works well for the fixed-camera setup. YOLO would be overkill for binary piece-present detection.

What this is not

• Not a general-purpose part feeder. The C-channel design assumes LEGO-sized pieces (~50mm sphere max). Industrial bowl feeders handle a broader size range.
• Not vibration-free in all configurations. The feeder upstream of the C-channels may still use vibration to move bulk LEGO into the first stage.
• Not a published research result. This is empirical engineering — the design emerged from iterative prototyping, not from simulation or academic study.

Where to go next

• Feeder experiments — the turntable and vibration approaches that preceded C-channels
• Object detection research — the detector models that find pieces inside the classification chamber
• Sorter architecture — how the C-channel control loop fits into the overall software