Warehouse AMR

The deployment context

Warehouse AMRs share space with human workers, forklifts, pallet jacks, and stationary racks that reconfigure on quarterly cycles. The floor is flat, structured, and well-lit — favorable for LiDAR-based localization. The challenge is dynamic: pedestrian foot traffic at intersections, forklifts reversing without warning signals, and temporary pallet stacks that appear overnight and block mapped corridors.

PathVynt's Balanced config tier was designed specifically for this profile. The fusion engine maintains a 5–10cm occupancy grid at 20Hz — fine enough to catch a pedestrian turning a corner into the AMR's path with 800ms of reaction time at typical AMR speeds (0.8–1.2 m/s).

Key configuration choices

• Sensor config: 16-channel mid-range LiDAR + monocular camera. Single LiDAR is sufficient for indoor rack geometry; mono camera provides pedestrian classification signal without stereo compute overhead.
• Drift model: FusionConfig.drift_model = WAREHOUSE_INDOOR — optimized for AMR vibration profiles on smooth concrete floors.
• Prediction horizon: 0.5s primary, 1.0s secondary. Forklift lateral speed rarely exceeds 0.5 m/s turning, so 1.0s is the critical horizon for corridor intersection risk scoring.
• Occupancy resolution: 5cm in the immediate zone (0–3m), 10cm in the extended zone (3–8m). Higher resolution near the robot, coarser farther out — saves compute without sacrificing safety-critical detection.

FusionConfig cfg;
cfg.lidar_model   = LiDARModel::MID_16;
cfg.camera_mode   = CameraMode::MONO;
cfg.drift_model   = DriftModel::WAREHOUSE_INDOOR;
cfg.occ_resolution = 0.05f;   // 5cm near zone
cfg.pred_horizons  = {0.5f, 1.0f};

PathVyntRuntime rt(cfg, "warehouse_map_v3.pvmap");
rt.start();