Warehouse Robot Sensors: What Keeps an AMR Localized and Safe

Localization answers “where am I?” Obstacle perception answers “what is around me?” Functional safety answers “must I slow or stop to keep risk within the validated limits?” Inventory sensing answers “which item or location is this?” A single camera or LiDAR may contribute to several tasks, but its safety rating and failure diagnostics determine whether it can be trusted for a protective function.

This distinction prevents a common mistake: assuming a robot with impressive 3D perception automatically satisfies the safety function needed for people sharing the aisle.

LiDAR measures range by sending laser light and timing or otherwise estimating the return. A mobile robot can use range scans for map matching, obstacle detection and localization. A safety laser scanner is designed and certified for protective functions, with monitored fields and safety outputs integrated into the robot safety architecture.

MiR lists two front/rear SICK nanoScan3 safety scanners on the MiR250, providing 360-degree protective coverage. The important design question is scanner placement: pallet overhang, forks, low objects and elevated obstacles can sit outside a single horizontal plane.

Depth cameras provide a three-dimensional point representation useful for pallets, rack geometry and obstacles above or below a planar laser. MiR250 lists two 3D cameras with a defined field of view for pallet and obstacle detection.

Camera performance changes with lighting, transparent or reflective materials, dark surfaces, dust and occlusion. Test the actual shrink wrap, black plastic, metal racks and floor reflections in the facility rather than relying on a clean demo environment.

Wheel encoders estimate distance from wheel rotation. An IMU measures angular velocity and acceleration. Combined, they give a fast local motion estimate between map updates, but slip, wheel wear and uneven flooring create error.

Warehouses with ramps, polished floors or debris should test odometry under the heaviest expected load. A localization stack must detect when predicted motion and external observations diverge rather than silently accumulating pose error.

Ultrasonic, infrared, short-range time-of-flight and contact sensors can cover near-field gaps that larger sensors miss. Forklift-style AMRs may also use fork-height, mast, load-presence or pallet-position sensing. The exact stack depends on the load-handling mechanism.

For docking, a local sensor or marker can be more useful than a globally accurate map. The robot may know its position to centimeters in the aisle but need a tighter final alignment to transfer a pallet or mate with a conveyor.

Barcode scanners, QR readers, RFID and cameras connect movement to inventory identity. An inventory-scanning robot may use high-resolution cameras or RFID antennas while its base uses a separate navigation and safety stack.

The business failure mode is different: a navigation error can stop or collide; an identification error can silently corrupt stock records. Measure read rate, false associations and recovery workflow separately from driving performance.

Ask what the robot does if a camera is blinded, a scanner window is dirty, an encoder fails, a localization confidence score drops or a payload blocks a field of view. A safe degraded mode may reduce speed, stop, request human assistance or switch to a redundant reference.

Sensor fusion is not magic redundancy. Two algorithms built on the same obstructed camera can fail together. True robustness comes from diversity of sensing, diagnostics and an explicit state machine for degraded operation.