Autonomous Obstacle Avoidance for UAVs: Algorithms and In-Depth Case Studies

Autonomous obstacle avoidance is crucial for safe and efficient UAVs (unmanned aerial vehicles) operation in complex low-altitude environments in a wide range of applications, including logistics, infrastructure inspection, agriculture and emergency response. In this paper, some of the prominent UAV obstacle avoidance methods such as A* and D* search algorithms, artificial potential field methods, deep reinforcement learning and multi-sensor fusion are reviewed. Their fundamental concepts, common enhancements, advantages and disadvantages and use cases are contrasted. Search-based methods are good at giving an explanation for the planning process and the plans are reliable globally, but they can be computationally expensive and do not perform well in real time. Artificial potential field methods are efficient in computation but the problems of local minima and unreachable target exist. Deep reinforcement learning is good at adapting to changing environments, but has problems with sample efficiency and generalization. Multi-sensor fusion provides complementary information for perception, localization and mapping that increases robustness. Future research should focus on the integration of the models and data, lightweight multimodal perception, and safe and efficient cooperative obstacle avoidance for multiple UAVs. These achievements are likely to make these systems more robust to uncertainty and facilitate wider deployment in the real world.