Skip to content

Freespace Reeds-Shepp planner that is an order of magnitude faster than OMPL's implementation with no compromise, no parallelization, and no multi-threading.

License

Notifications You must be signed in to change notification settings

IbrahimSquared/accelerated-RS-planner

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

50 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Accelerated Reeds-Shepp Planning

Solves the Reeds-Shepp path planning problem 15x faster on average than OMPL's implementation.

Paper attachment TBD.

Reduces set of cases to consider down to 20 and introduces a new partition that speeds up the computation of the shortest path.

Requires SFML & OMPL. It has a demo in main.cpp that computes and draws paths between randomly generated start/end configurations.

Also contains an implemntation of "An efficient algorithm to find a shortest path for a car-like robot".

Benchmarking Results

This table presents sample benchmarking results of our proposed planner against OMPL's implementation of the original Reeds-Shepp algorithm and against our implementation of Desaulniers1995.

Performed on idle Linux Intel i9-13980HX.

Method Average Time per State (µs) Time Ratio to OMPL Maximum Path Length Error Average Path Length Error
Proposed 0.0827 15.12 9.82e-15 3.28e-16
Desaulniers1995 0.216 5.79 5.82e-15 2.77e-16
OMPL 1.25 1 - -

Table: Benchmarking results of our proposed planner against OMPL's implementation of the original Reeds-Shepp algorithm and against our implementation of Desaulniers1995.

Sample partition in 3D

3D partition of regions in the configuration space. Each color refers to a unique region (out of 20) where one identified path type is certainly the shortest. 1 million configurations $p_{f}$ spanning $\big[ [-100,100,], [-100,100], [-\pi,\pi)\big]$ for $p_{0} = (0,0,0)$ and $r = 20$. Red axis = $x$ axis, blue axis = $y$ axis, yellow axis = $\theta$ axis.
alt text

Variability Based on Types

1e8 final configuration samples spanning $\big[ [-1000,1000,], [-1000,1000], [-\pi,\pi)\big]$ for $p_{0} = (0,0,0)$ and $r = 400$. Computing those speedups for each path type incurs an overhead that makes the speedups slightly slower. This is only to show that the speedup is not uniform across all path types. The weighted average speedup is around 11.87.

Type Occurrences Speedup
1 6829417 11.5399
2 8684623 13.1266
3 9879506 12.3335
4 7630623 13.4698
5 6732552 12.8536
6 6683699 11.8223
7 791193 11.4474
8 2287070 11.8638
9 554434 11.7778
10 2529318 11.9787
11 1985832 10.639
12 11555 10.8744
13 7494149 10.3889
14 8013981 11.3507
15 7714423 9.52923
16 9776168 11.3254
17 1118135 10.2761
18 9897533 11.0233
19 138503 9.18825
20 1247286 8.42923

How to Use (to be completed)

Set compiler path if needed. Make sure SFML:
sudo apt install libsfml-dev
and OMPL libraries:
https://ompl.kavrakilab.org/installation.html
are installed, then:
sudo apt install cmake
sudo apt install g++
mkdir build && cd build
cmake ..
make

Troubleshooting build issues with OMPL:
By default, OMPL installs in /usr/local/include/ompl-1.6/ompl. Create a symbolic link so that it can be properly located:
sudo ln -s /usr/local/include/ompl-1.6/ompl /usr/local/include/ompl
Similarly, create Eigen symbolic links so that required files can be found by OMPL: cd /usr/include
sudo ln -sf eigen3/Eigen Eigen
sudo ln -sf eigen3/unsupported unsupported
After making any changes to the library paths or creating symbolic links, run the ldconfig command to update the system's cache of shared libraries:
sudo ldconfig
After building, you can verify that the required libraries have been properly found using ldd, for example:
ldd ./acceleratedRSPlanner

To use this project, follow these simple steps:

About

Freespace Reeds-Shepp planner that is an order of magnitude faster than OMPL's implementation with no compromise, no parallelization, and no multi-threading.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published