The so-called path planning is to find a continuous motion trajectory between the starting point and the end point, and avoid obstacles in the environment while optimizing the path as much as possible.
Commonly used path planning algorithms include traditional graph-based search algorithms, sampling-based path planning algorithms, and path planning algorithms considering dynamics. So, under what circumstances are these path planning algorithms applicable? Next, we will give you a detailed introduction.
>>>>
Search-Based Path Planning Algorithm
Search-based path rule algorithms mainly include: BFS (breadth-first) algorithm, DFS (depth-first) algorithm, Dijkstra algorithm, heuristic search A* algorithm, etc.
In general, search-based planning algorithms are suitable for running on raster maps . By continuously searching on the grid map, a continuous trajectory to the destination is retrieved .
Although the planning algorithm based on graph search can always give a global optimal solution (the shortest path and the best efficiency), when the map is too large and the planning dimension is too high, its search efficiency will become very slow.
>>>>
Sampling-Based Path Planning
Sampling-based path planning mainly includes: RRT (rapidly expanded random tree) algorithm, RRT* algorithm, informed RRT* algorithm, etc.
In some scenarios, the focus of the path planning algorithm is mainly on efficiency , so the sampling-based planning algorithm is more in line with the requirements.
The core of this type of algorithm is random sampling . Starting from the parent node, randomly generate child nodes on the map, connect the parent and child nodes and perform collision detection. If there is no collision, expand the child node. By continuously expanding the sample points randomly until a path connecting the starting point and the ending point is generated . Since the sample points are randomly generated, the final execution path may not be the global optimal solution , and it may even be obvious that the robot is "detouring".
Compared with the search-based planning algorithm, the sampling-based planning algorithm is more efficient because it can quickly generate a feasible path without traversing the entire grid map.
The above two types of algorithms do not take into account the robot motion constraints , but only consider the "closest path" or "fastest path". If you want to obtain the global optimal solution in the shortest time , then a path planning algorithm that considers dynamics will be a good choice.
>>>>
Path Planning Considering Dynamics
Path planning considering dynamics mainly includes: hybrid A* algorithm, Kinodynamic RRT* algorithm, etc.
Taking an unmanned vehicle as an example, if a path with a right-angle turning point is generated, the unmanned vehicle of the two-wheel differential motion model can barely pass the right-angle turning point, because the minimum turning radius of the differential model unmanned vehicle is 0, which means It can rotate in place (but it can't achieve instant steering in the actual movement process). For the unmanned vehicle of the Ackerman motion model (the car in life is the Ackermann steering system), since the turning radius is not 0, it cannot pass through the right-angle corner. The planning algorithm considering dynamics no longer considers the robot as a particle, but considers that the planned trajectory must satisfy the dynamics , and the generated trajectory can make the robot actually execute .
Taking the hybrid A* algorithm as an example, after considering the vehicle kinematics, the resulting path is not a straight line, but a smoother curve, and when detecting obstacles, the robot is not simply regarded as a mass point, but The robot profile is taken into account. As shown in the figure below, when the A* algorithm expands the node, there are eight grids. When the Hybrid A* algorithm expands the node, there are six motion primitives, three motion modes forward and three motion modes backward.
