Perception Layer
Camera frame → Supabase Edge Function → Roboflow Hosted API → detection JSON → robot command.
CAM-01 · Infeed Conveyor1920×1080 · 30fps · STANDARD
frame 1247
case · 98.4%
rot +12° · pickability high
case · 96.9%
rot -8° · pickability high
● TRACKING 2 OBJECTS
latency 42ms
model: polyborg-detect-v3.2
▶ ROBOT ACTION QUEUED
generate_pick_path · vacuum gripper
Detection Results
Normalized output from run-roboflow-inference
| class | conf | x | y | w | h | rot° | pickability |
|---|---|---|---|---|---|---|---|
| case | 98.4% | 400 | 280 | 240 | 220 | 12 | high |
| case | 96.9% | 980 | 320 | 240 | 220 | -8 | high |
Robot Action Preview
High-confidence case detected — pick command queued in robot_commands.
{
"next_step": "generate_pick_path",
"gripper": "vacuum",
"target": {
"class": "case",
"confidence": 0.984,
"x": 400,
"y": 280,
"rotation": 12
},
"status": "queued"
}Refractive Vision Mode
Improves detection reliability for clear plastic, glossy labels, reflective packaging, and difficult lighting.
Detection Confidence
98.4%
Inference Time
42ms
round-trip
Objects Detected
2
Pose Accuracy
96.1%
Recovery Events
4
New Examples Logged
1,284
for fleet learning
Detection Event Stream
Frame-by-frame perception log
→ Frame 1242: Case detected, 98.4%
→ Frame 1243: Pose estimated
→ Frame 1244: Gripper target selected
→ Frame 1245: Pick path generated
→ Frame 1246: Command queued
→ Frame 1247: Detection logged