Falcon: Random Network Distillation Enhanced Energy-Efficient Scheduling in Heterogeneous UAV-Enabled Edge Computing

UAV-enabled edge computing offers dynamic service provisioning in scenarios where terrestrial infrastructure is overloaded or inaccessible. However, efficient workload scheduling in such hybrid systems is challenged by resource heterogeneity and stringent UAV energy constraints. To address these issues, we propose Falcon, a reinforcement learning-based scheduling framework that integrates Random Network Distillation (RND) with Proximal Policy Optimization (PPO) to enhance exploration in sparse-reward environments. The scheduling problem is formulated as a Markov Decision Process (MDP) with a composite reward function balancing resource utilization, load balancing, and energy efficiency. Intrinsic rewards from RND accelerate convergence by encouraging exploration of under-explored states, which reduce convergence time by 9.5%. Experiments on realworld workload traces from the Alibaba Cluster Trace Dataset demonstrate that Falcon outperforms k8s scheduling methods in resource utilization by 6.7% and energy consumption by 14.8%.