The Corridor Map Method: Real-Time High-Quality Path Planning

1
The Corridor Map MethodReal-Time High-Quality
Path Planning
Roland Geraerts and Mark Overmars ICRA 2007
2
Previous work

• Potential field planners
• Flexible
• Slow / local minima
• Probabilistic Roadmap Methods
• Fast
• Ugly paths
• Output fixed paths in response to a query
• Predictable motions
• Lacks flexibility when environment changes

3
Our planner

• Requirements
• High-quality paths
• Flexible
• Extremely fast
• Current limitations
• The robot is modeled by a disc
• Experiments with only 2D problems

4
The Corridor Map Method

• Construction phase (off-line)
• Create a system of collision-free corridors for
  the static obstacles

Graph
Corridor map graph clearance
5
The Corridor Map Method

• Query phase (on-line)
• Extract corridor for given start and goal
• Extract path by following attraction point

6
The Corridor Map Method

• Attraction point a(x)
• Robots location x
• Robots goal g
• Radius circle r
• Euclidean distance d
• Path is obtained by integration over time while
  updating the velocity, position, and attraction
  point of the robot
• For other behavior locally adjust robots path
  by adding forces

7
Avoiding obstacles

• Adding forces
• For each obstacle, add repulsive force to the
  robot

• Creating a sub-corridor
• For each obstacle, move backbone path locally and
  recompute clearance info

8
Creating shorter paths

• Attraction point a(x) corresponds to point Bt
  on the backbone path
• Add additional valid attraction point a(x, ?t),
  corresponding to point Bt ?t
• Valid means x can see point Bt ?t

9
Experimental setup

• Single path planning system
• Created in Visual C, Windows XP
• 2.66 GHz P4 processor, 1 GB memory
• Each experiment was run 100 times
• Statistics running time of query phase, CPU load
• Input graphs created using
• Creating High-quality Roadmaps for Motion
  Planning in Virtual Environments- IROS 2006
• Environments were discretized 100x100 cells

10
Experimental setup

• Maze

• Field

1.6 seconds
20 seconds
11
Experiments Smooth paths

• Maze
• Query time 2.41 ms
• CPU load 0.026

• Field
• Query time 0.84 ms
• CPU load 0.029

12
Experiments Obstacles

• Maze adding forces
• Query time 7.09.0 ms
• CPU load 0.050.06

• Maze sub-corridor
• Query time 3.013.6 ms
• CPU load 0.0250.10

13
Experiments Obstacles

• Maze

14
Experiments Obstacles

• Field adding forces
• Query time 2.02.3 ms
• CPU load 0.050.05

• Field sub-corridor
• Query time 1.07.0 ms
• CPU load 0.030.16

15
Experiments Obstacles

• Field

16
Experiments Short paths

• Maze ?t 0
• Query time 2.41 ms
• CPU load 0.026

• Maze ?t 0.2
• Query time 9.64 ms
• CPU load 0.104

17
Experiments Short paths

• Maze

18
Experiments Short paths

• Field ?t 0
• Query time 0.84 ms
• CPU load 0.029

• Field ?t 0.2
• Query time 3.36 ms
• CPU load 0.116

19
Experiments Short paths

• Field

20
Conclusions

• The CMM produces high-quality paths
• Natural paths smooth, short / large clearance
• The CMM is flexible
• Paths are locally adjustable
• The CMM is fast
• CPU load lt 0.1

21
Future work

• Extend experimentswith 2½D / 3D problems
• Study applications
• Planning of a group
• Steering a camera
• Alternative routes
• Tactical planning