基于采样的路径规划——深蓝学院Lecture3

0 Preliminaries

0.1 From workspace to configuration space

• Robot transformed into a point in configuration space
• Paths transformed from swept volume into 1 dimensional curves.

0.2 Sampling-based Planners (or Probabilistic Methods)

+ Capture the connectivity of the solution space to bring a feasible or optimal solution by constructing Trees or Graphs connected by discrete C-space samples sampled incrementally or in batch in the continuous C-space without explicitly constructing it.
+ Their performance mostly depends on sampling-strategy, collision-checking and neighbor-search.

0.3 Two fundamental Tasks

• Exploration
  Acquires information about the topology of the search space, i.e. how subsets of the space are connected.
• Exploitation
  Incrementally improves the solution by processing the available information computed by the exploration task.

0.4 Terminology

• Probabilistic Completeness (概率完备)
  A feasible solution will be found, if one exists, with probability one as the number of samples approaches infinity.
• Asymptotical(near) Optimality(渐近最优)
  The cost of the returned solution converges almost surely to the optimum as the number of samples approaches infinity,
• Anytime (即时)
  Quickly find a feasibe but not necessarily optimal solution, then incrementally improve it over time.

0.5 Brief History

PRM 1996 TRA
RRT 1998
RRT* 2011 IJRR
RRT# 2013 ICRA
Informed-RRT* 2014 IROS
GuILD 2021 Arxiv

1 Feasible path planning methods

1.1 Probabilistic Road Map (PRM)

What is PRM?

• Probe(探究) space connectivity in a graph structure,
• fine converage of free space
• Divide planning into two phases:
  + Learning phase
  + Query phase
  Learning phase:
  Detet the C-space with random points and construct a graph that represents the connectivity of the environment.
  • Sample N points in C-space
  • Delete points that are not collision-free
  • Connect to nearest points and get collision-free segments
  • Delete segments that are not collision-free
    Query phase:
    Search on the road map to find a path from the start to the goal (using Dijkstra’s algorithm or the A* algorithm)
  • Road map is now similar with the grid map (or a simplified grin map).
    Pros: can conduct multiple queries (static map).
    Cons: Build graph over state space but no particular focus on generating a path.
    Improving efficiency
    The collision-checking process is time-consuming, especially in complex or high-dimensional environments.
• Lazy collision-checking
  check collisions only if necessary

1.2 Rapidly-exploring Random Trees(RRT)

Build up a tree from start to goal through generating “next states” in the tree by executing random controls

2 Optimal path planning methods

Optimal Criterion
Dynamic Programming(DP) functional equation (discrete form)
Direct DP Methods

2.1 RRT*

3 Accelerate convergence

4.1 RRT#,