Automated Tracking of Pallets in Warehouses: Beacon Layout and Asymmetric Ultrasound Observation Models

We consider the use of wireless sensor networks to automatically track "perceptive pallets" of materials in warehouses for the purpose of monitoring volumetric and spatial constraints. A combination of radio frequency and ultrasound chirping produces position estimates that are noisy and prone to error. To address this, we measure and characterize the ultrasound response from standard "Cricket" wireless sensor motes and beacons. We develop a non-parametric particle filtering approach to estimate trajectories of moving motes and introduce two asymmetric observation models that incorporate measured cardioid-shaped response patterns of ultrasound. We use simulation to study the effects of mote placement: position error as a function of ceiling height and beacon density, and then perform physical experiments to evaluate the effectiveness of asymmetric vs. symmetric observation models for pallet tracking. Experiments suggest that asymmetric observation models can improve position estimates by as much as 11%.