Recent Radio-Frequency Identification (RFID)-Enabled Autonomous Logistics Management (REALM) advances in beta versions of a novel deep learning tool, “Tagnet”, and a random forest classification tool, “Fangorn”, led the crew and ISS Inventory Stowage Officers (ISOs) to a missing Cargo Transfer Bag (CTB) in Node 4. Tagnet was conceived by EV8 intern Joel Simonoff during a summer tour in 2019, and continued in 2020. Although further planned testing is required, this early success was an exciting step in the application of machine learning applied to RFID. The CTB was reported lost approximately a year ago, but was not required by the crew until April 20th, 2020. Initial human inspection of raw data indicated that the missing CTB was likely in Node 1 or PMA 1, but localizing to the rack level had proven exceedingly difficult using human analysis and other deterministic and machine learning tools to date. To further complicate localization of this particular CTB, the two readers in Node 1 had not been properly reading tags due to a network anomaly that is being worked. Tagnet and Fangorn were able to localize the CTB to the Node 4 location using previous training on several TB of archived and new data obtained through only US LAB RFID readers. Data mining and derivation of machine inferences from raw RFID data is a critical technology focus of the REALM project, as RFID technology has not previously been applied to localizing items to the level of accuracy required by NASA. Moreover, the complex scattering environment faced by NASA spacecraft and habitats presents additional RFID localization challenges. – Patrick Fink, EV8 REALM Deep Learning Directs Crew to Missing CTB on ISS
