Efficient Sampling-based Multirotors Kinodynamic Planning with Fast Regional Optimization and Post Refining

Position Checking-Based Sampling Approach Combined with Attraction Point Local Optimization for Safe Flight of UAVs

“…For the convenience of description, as shown in Figure 9, the trajectory color of both algorithms is uniformly set to red, the trajectories around the vertices are marked by blue dashed boxes, and the guided graph is indicated by green lines. Since the spherical space for position checking of the state points is delineated around the vertices, it can be seen in Figure 9 that the improved algorithm in this paper generates trajectories further away from the obstacles around the vertices, as compared to [10]. This indicates further that the algorithm proposed in this paper is able to reduce the risk of UAV flight around obstacles in order to ensure the safe flight of UAV.…”

“…This process is performed in Expansion (x new , X backward nearest ) in Algorithm 1, which ensures that other optimal or suboptimal sampling points are not affected and can be discarded before the two-point BVP is solved. In contrast to the algorithm in [10], a spherical space divided around the vertices of the topology-guided graph is proposed in the paper to perform a position check on the newly sampled points. This additional position checking not only reduces the number of triggers for local optimization of the back-end trajectory but also makes the flight safer.…”

“…It is easy to see from the above Figures 4 and 5 that, since the constructed environment is very rough, relying only on this original trajectory for generating the final reference trajectory will result in a degradation of the quality of the generated trajectory. Therefore, a local optimization triggered by collision detection is proposed in [10] where, instead of direct trajectory discarding when a collision occurs in the generated trajectory, the failing trajectory is reused and locally adjusted by a gradient-based optimization scheme.…”

“…where C att denotes the set of attractive points, and attSeg denotes the trajectory of the corresponding segment. Different attraction points are selected by the difference of environment, meaning that therefore the algorithm proposed in this paper is able to achieve better results in cluttered environments, as compared to [10]. In addition, Algorithm 3 gives the detailed process.…”

“…Firstly, the topology of the environment is constructed through start and end points; next, the dynamics are sampled at the periphery of topological structure; and finally, the reference trajectory is obtained through local optimization and adjustment, which largely reduces the number of sampling points. In order to avoid the case that some narrow slits in the environment may provide more optimal solutions, Ye et al, in [10], propose a sampling-based kinodynamic planning method combined with local optimization, but this increases the risk of UAV collisions.…”