UCLA

Artificial intelligence (AI) systems typically do not have the ability to execute inference and learning algorithms concurrently in real-time like the human brain. Instead, they typically execute these algorithms in series, which can be inefficient and lack adaptability in complex and unpredictable real-world situations. In this work, we present an intelligent system that combines a drone and a synaptic resistor (synstor) circuit to concurrently execute inference and reinforcement learning algorithms in real time. The synstor circuit's conductance matrix is able to adapt and optimize in real-time learning processes, allowing the drone to navigate towards its target positions even in chaotic aerodynamic conditions. In learning experiments involving synstor circuits, human neurobiological circuits, and artificial neural network (ANN), the synstor circuit's real-time learning outperformed both human real-time learning and ANN’s iterative offline learning process in terms of adaptability, learning speed, accuracy, power consumption, and energy efficiency. The use of synstor circuits to bypass the limitations of traditional computers offers the potential for the development of AI systems with brain-like real-time learning abilities, high energy efficiency, and adaptability in complex real-world environments for a wide range of applications.