A Few-shot Dynamic Obstacle Avoidance Strategy in Unknown Environments

Obstacle avoidance is one of the basic capabilities of intelligent mobile robots. With the diversification of the application environment, mobile robots are required to avoid obstacles with higher generality. Benefit from the development of mobile platform and deep learning algorithm in recent years, we conceive a few-shot dynamic obstacle avoidance strategy to meet this higher generality demand. Under this metric-based metalearning method, mobile robots can quickly adapt to unknown environments by learning from several samples. In order to verify its effectiveness, we use this strategy to train a model and deploy it to the mobile robot and run multiple obstacle avoidance recognition tests in the real-world environment. The results of experiments performed on the mobile robot platform illustrates a good performance and verifies our proposed strategy. In addition to analyzing the experimental results, the advantages, disadvantages as well as application potential of the proposed strategy as a decision aid are also discussed.