This paper mainly combines the perception and decision-making of aerial robots to carry out the following research. Unstructured environment perception and modeling problems: The first problem is perception. The high-speed response characteristics of drones determine that their environmental perception poses a higher challenge to existing methods. At the same time, the non-conformity of non-structural environments description and time-varying features also make traditional modeling methods fail in this environment. In order to perceive the environment more quickly, this paper proposes a method based on loop detection compensation to achieve high-speed sensing. Based on perception, this paper proposes a discrete Gaussian model modeling method to characterize the non-structural environment. The path planning and motion planning of the aerial robot: Aerial robot requires a faster planning response to complete the adaptation to the dynamic environment, and as the environment becomes more complex, the motion planning space of the drone is usually faced with dealing with a large number of obstacles. This often requires algorithms to be able to perform computationally obstacle collision detection and safe area monitoring. The traditional algorithm is very time-consuming for the processing of unstructured narrow channel environments. This paper proposes a random sampling algorithm based on environmental information, combined with the dual information induction from obstacles and targets so that the algorithm can effectively generate path links in narrow channel areas to complete fast path planning. At the same time, this paper also proposes to control the smoothing motion trajectory based on the sixth-order Bezier curve, so that the flying robot can conduct online navigation with minimum energy and risk coefficient. Online dynamic threat avoidance of aerial robots: For robots, the modeling and avoidance of dynamic obstacles is one of the difficult points, especially when performing high maneuvering movements. Especially, if the navigation planner needs a very fast response. it requires the planner to have a high speed and predictive ability to evade. This paper solves the multi-path search problem of a random search tree for the first time and proposes a fast maneuver obstacle avoidance based on path planning. This method has been experimentally proven to have better timeliness than the traditional re-planning direction, and it is also possible to dynamically return to the previous trajectory to avoid the problem of reverse motion obstacles. Finally, in the conclusion and prospects, the research on the thesis is summarized, and the possible future research directions are briefly stated.