Experimental Demonstration of Temporally Aware Fault‐Tolerant Sensor Fusion Using Memristive Associative Learning

Autonomous vehicles rely on fault‐tolerant multi‐sensor fusion for accurate perception and reliable decision‐making in case of sensor dropouts. However, conventional fault‐tolerant mechanisms struggle to fully acknowledge the valid sensor pairs (or a set) associated with individual events, often producing erroneous activations when signals are delayed, misaligned, or partially available. This work presents a mixed‐signal memristive associative learning circuit (ALC) that introduces two key mechanisms, selective sensor pairing and electronically tunable temporal validation, to ensure that only correctly grouped and temporally coherent sensor inputs contribute to learning and output generation. Implemented using a memristor, operational amplifiers, logic gates, and latches, the proposed design consumes less than 300 mW with highly‐scalable integrability in standard CMOS technologies. Simulations show that the circuit reduces erroneous activations from 78.6% to 0% when compared with the traditional associative learning realization. Finally, the proposed proof‐of‐concept is experimentally validated based on a benchtop implementation using commercial memristor IC as well as memristor emulator circuit and results are compared. Additionally, we demonstrate the potential use of ALC output in real‐time scenarios through a servo control, i.e., how the proposed architecture can drive the actuators in real vehicles, showing the efficacy of the idea in dealing with real‐time responses.