Preventing False Activations in Autonomous Vehicles: A Memristive Associative Learning Approach with Selective Sensor Pairing

Abstract

Autonomous vehicles rely on multi-sensor fusion for accurate perception and decision-making. However, conventional sensor-based learning circuits struggle to differentiate between incomplete and valid sensor inputs, leading to erroneous activations. This paper presents a memristive associative learning circuit with selective sensor pairing and temporal validation to enhance fault tolerance in autonomous driving. Unlike existing associative learning circuits that generate outputs in untrained states with partial sensor data, the proposed design enforces a strict, electronically adjustable temporal window between sensor inputs before triggering an output. This mechanism prevents false activations from isolated or delayed signals, ensuring decisions are based on complete situational awareness. The circuit implemented using a memristor, operational amplifiers (OP-AMPs), logic gates, and latches, achieves power consumption below 300 mW, making it a low-power and efficient solution. Simulation results show that the proposed circuit reduces erroneous activations by 60%, achieving an average error rate of 0.96%, compared to 30.98% in a traditional associative learning circuit during pedestrian detection scenarios. Our finding demonstrates that selective sensor pairing, and temporal validation significantly improve the reliability and safety of autonomous vehicle navigation, particularly in challenging environments where sensor signals may be delayed or partially available.