Fully Spiking Actor-Critic Neural Network for Robotic Manipulation

This study proposes a hybrid curriculum reinforcement learning (CRL) framework based on a fully spiking neural network (SNN) for 9-degree-of-freedom robotic arms performing target reaching and grasping tasks. To reduce network complexity and inference latency, the SNN architecture is simplified to include only an input and an output layer, which shows strong potential for resource-constrained environments. Leveraging SNNs for their high inference speed, low energy consumption, and biological plausibility grounded in spikes, we integrate a curriculum that partitions training by temporal progress with Proximal Policy Optimization (PPO). Meanwhile, an energy consumption modeling framework is introduced to quantitatively compare the theoretical energy consumption between SNNs and conventional Artificial Neural Networks (ANNs). A dynamic two-stage reward adjustment mechanism and optimized observation space further improve learning efficiency and policy accuracy. Experiments on the Isaac Gym simulation platform demonstrate that the proposed method achieves superior performance under realistic physical constraints. Comparative evaluations with conventional PPO and ANN baselines validate the scalability and energy efficiency of the proposed approach in dynamic robotic manipulation tasks.