This work integrates concurrent team learning into a deep reinforcement learning model for collaborative multiagent problems to reach a common goal. RL trains each agent to decide its best actions, and then the joint decision is made to accomplish the assigned team task based on the collaborative platform. Each agent learns individually and shares the observed feature with others. The collaborative platform makes the decision based on these features provided by agents. The collaborative platform consists of two attention mechanisms—soft and self-attention. The self-attention module is applied to distinguish local environmental features sensed by an individual agent for concurrent learning. The platform integrates collective information locally from agents through a soft attention interface. A termination network is introduced to the collaborative model to determine the number of elapsed sequences in collecting sufficient information to support the joint decision instead of a fixed number of iterations. Experiments on limit-sighted agent deployment for static image classifications, video game playing, and aerial image object detection of multiple airborne sensors have evaluated the performance of the proposed model. The experimental results show that the proposed model provides a more efficient collaborative mechanism for multiagent systems than the state-of-the-art methods.