Obstacle Avoidance Algorithm via Hierarchical Interaction Deep Reinforcement Learning

The navigation task in a complex scenario is an essential problem in mobile robot technology. The mobile robot obstacle avoidance algorithm plays a vital role in navigation. In the navigation task, the mobile robot has to select the optimal action under different conditions in real-time. This research proposes a novel obstacle avoidance algorithm based on deep reinforcement learning. The proposed algorithm utilizes interacting with the environment in the simulation to update the decision network. The decision network includes the feature extraction module and the hierarchical interaction module. The feature extraction module can extract and identify the features of dynamic obstacles in the scenario. And the hierarchical interaction module can handle the interaction features between the mobile robot and obstacles. Furthermore, a safety module is applied in the algorithm to guarantee mobile robot collision-free. Finally, the experiment is conducted to evaluate the proposed method in the simulation environment. The experiment result verified the safety and effectiveness of the proposed method and proved that the proposed method could ensure the mobile robot completes the task.