Autonomous modeling of artificial swarms is necessary because manual creation
is a time intensive and complicated procedure which makes it impractical. An
autonomous approach employing deep reinforcement learning is presented in this
study for swarm navigation. In this approach, complex 3D environments with
static and dynamic obstacles and resistive forces (like linear drag, angular
drag, and gravity) are modeled to track multiple dynamic targets. Moreover,
reward functions for robust swarm formation and target tracking are devised for
learning complex swarm behaviors. Since the number of agents is not fixed and
has only the partial observance of the environment, swarm formation and
navigation become challenging. In this regard, the proposed strategy consists
of three main phases to tackle the aforementioned challenges: 1) A methodology
for dynamic swarm management, 2) Avoiding obstacles, Finding the shortest path
towards the targets, 3) Tracking the targets and Island modeling. The dynamic
swarm management phase translates basic sensory input to high level commands to
enhance swarm navigation and decentralized setup while maintaining the swarms
size fluctuations. While, in the island modeling, the swarm can split into
individual subswarms according to the number of targets, conversely, these
subswarms may join to form a single huge swarm, giving the swarm ability to
track multiple targets. Customized state of the art policy based deep
reinforcement learning algorithms are employed to achieve significant results.
The promising results show that our proposed strategy enhances swarm navigation
and can track multiple static and dynamic targets in complex dynamic
environments.

本研究提出一种基于深度强化学习的自主建模人工群体的方法，以解决手动创建过程时间复杂且不切实际的问题。该方法通过对静态和动态障碍物及阻力的建模，实现了多目标追踪和鲁棒群体形成，而采用的策略包括动态群体管理、避免障碍与追踪对象及岛屿建模。创新的基于策略的深度强化学习算法取得了显著的结果，表明该方法在复杂环境下显著提高了群体导航和追踪多个目标的能力。