Hardware implementation of photonic neuromorphic autonomous navigation

Abstract

Reinforcement learning enables artificial intelligence to move beyond perception toward decision-making, but its deployment on conventional electronic hardware is limited by the latency and energy consumption of the von Neumann architecture. Here, we propose a photonic spiking twin delayed deep deterministic policy gradient reinforcement learning architecture for neuromorphic autonomous navigation and experimentally validate part of the architecture through hardware-software co-inference using a distributed feedback laser with a saturable absorber (DFB-SA) array. The architecture integrates a photonic spiking Actor network with dual continuous-valued Critic networks, where the final nonlinear spiking activation layer of the Actor is deployed on the DFB-SA laser array. In autonomous navigation tasks, the system achieves an average reward of 58.22 ± 17.29 and a success rate of 80% ± 8.3%. Hardware-software co-inference demonstrates an estimated device-level nonlinear activation energy consumption of 0.78 nJ per activation event and a nonlinear activation latency of 191.20 ps under ideal parallel channel activation, with co-inference error rates of 0.051% and 0.059% for scenarios with and without obstacle interference, respectively. Simulations of error-activated channels agree well with the expected responses, validating the dynamic characteristics of the DFB-SA laser. The proposed architecture provides a promising pathway toward low-power, low-latency photonic neuromorphic autonomous navigation.