Markov Decision Processes: Discrete Stochastic Dynamic Programming
Markov Decision Processes: Discrete Stochastic Dynamic Programming
Antenna Theory: Analysis and Design
Antenna Theory: Analysis and Design
Planning and acting in partially observable stochastic domains
Artificial Intelligence
| Hi-index | 0.00 |
Inventory misplacement and inaccuracies contribute significantly to the operational expense of the overall supply chain. Radio Frequency Identification (RFID) technology has gained prominence as a solution to this approach. However, issues such as presence of metals and frequency interference hamper its performance and value derived out of its implementation in many environments. To analyze the true value of an RFID system we formulate a Partially Observable Markov Decision Process (POMDP) model for RFID directed search to detect misplaced items. A forklift operator (FLO) in a warehouse assigned to locate a misplaced item is guided by the imperfect variations in the strength of the signal received from the RFID tag (active or passive). The POMDP considers five possible FLO actions, five RFID observations in scenarios with different reader-tag signal strength distributions. The results indicate that the POMDP can provide shortest path to locate the tag when RFID signal strength is high. As the signal strength decreases, the signals received (observations) become more dispersed, and the number of steps to reach the tag increases considerably. The expected reward (roughly reduction in search effort) from a 20-step POMDP with high signal strength was 18 times higher than for the low strength (i.e., poor/dispersed observations) model. While even a poorly designed RFID system provides marginal benefits in environments devoid of a sophisticated warehouse management system (WMS), the results show that with the use of WMS (i.e., situations involving strong prior knowledge) adopting a poorly designed RFID system may actually cost the firm more than without it.