6G attempts to provide a network that seamlessly integrates terrestrial, air, and satellite networks. 6G3D network has proven to be highly advantageous in remote and rural areas, where terrestrial network deployment is complex or unfeasible due to network infrastructure constraints. However, such a 3D network is highly dynamic and resource-constrained, making task scheduling critical. Based on 6G's vision, the 3D network is to be operated with in-network intelligence, and many papers have proved that deep reinforcement learning (DRL) can improve task scheduling performance. However, they have not covered the online process for the on-demand or fast-deployed 3D network for 6G. This paper proposes an online DRL-based task scheduling algorithm, namely 3D-TS. The simulation results show that the DRL-based task scheduling algorithm improves scheduling quality by increasing training data in 3D networks while balancing scheduling quality and efficiency.