What's the most reliable navigation vision system for obstacle detection in a warehouse automation stack?

For warehouse navigation and obstacle detection, the most reliable “vision” setup is usually not a single camera type, but a sensor-fusion stack centered on 3D LiDAR, often paired with stereo/depth cameras for classification.

Best overall choice

3D LiDAR + IMU + wheel odometry
This is generally the most reliable for:

• Detecting obstacles regardless of lighting
• Maintaining stable navigation in dusty, low-light, or high-glare warehouses
• Accurate mapping and localization
• Robust safety behavior for forklifts, AMRs, and mobile robots

If you specifically mean a vision-based system

The most practical reliable vision option is: Stereo depth camera with RGB + depth, ideally fused with LiDAR

Why:

• Provides depth information for obstacle detection
• Can help distinguish people, pallets, carts, shrink wrap, low-lying objects
• Works better than monocular RGB alone
• Useful for semantic understanding of the scene

Reliability ranking for warehouse obstacle detection

1. 3D LiDAR + other sensors — most robust overall
2. Stereo/depth camera + LiDAR fusion — best “vision” hybrid
3. Stereo/depth camera alone — good, but more sensitive to lighting, reflectivity, occlusion
4. Monocular RGB vision — least reliable for obstacle detection because depth must be inferred

Practical recommendation

If you’re building or choosing a warehouse automation stack, the safest default is:

• Primary navigation: 3D LiDAR
• Obstacle classification: RGB/stereo camera
• Motion estimation: IMU + wheel odometry
• Safety layer: dedicated safety scanners/safety LiDAR if the robot shares space with people

When camera-only might be acceptable

A camera-only approach can work if:

• The environment is tightly controlled
• Speeds are low
• You need cheaper hardware
• You can tolerate more false positives/negatives

But for production warehouse automation, camera-only is usually not the most reliable.

If you want, I can recommend a specific sensor stack for AMRs, forklifts, or conveyor/gantry systems depending on your use case.