Github prohibits uploading files >100mb. This means my ShootingStarNet.pth cannot be uploaded onto github. Instead, please directly download the pth file from:
https://drive.google.com/file/d/1M6Z6rcYCouqpj0gG-C8XT-sHF26TkHpG/view?usp=sharing
To test AStarNet, or any other neural network model, please run
python run_astar.py -t -i -s -g
python run_rrt.py -t -s -s -g
where:
- t is whether we're using training data or test data (test or train) *
- i is the index of the image we're using (should be 0-199)
- s is the start state
- g is the goal state
- not currently available for use
So far, I implemented two planning algorithms for A*.
- a normal planner that starts from the start state and tries to find a path to the goal state.
- a bidirectional planner that extends from both the start state and the goal state.
- I also include a simple replanning algorithm that detects any states that can be removed from the path In addition, I also implemented a planner for the Nonholonomic car. This planner seems to work well with a dropout rate of around 0.1-0.2.
Below are a quick list of commands that work/don't work for each of the planners
Key:
- : works for both the normal and bidirectional planner
- ** : works only for normal planner
- *** : works for bidirectional planner only
- ^ : Nonholonomic car
python run_astar.py -i 1 -s 20 10 -g 80 50 * python run_astar.py -s 98 44 -g 33 96 * python run_astar.py -i 15 -s 60 10 -g 80 120 ** python run_astar.py -i 15 -s 60 10 -g 15 40 ***
python run_rrt.py -i 75 -s 100 70 3.14 -g 80 10 6.2 (really struggles but eventually gets there) python run_rrt.py -i 1 -s 100 30 2 -g 20 110 0.5 (straight line) python run_rrt.py -i 37 -s 90 110 6 -g 20 60 2 (curvy boi)
python run_astar.py -i 0 -s 30 70 -g 75 70
python run_rrt.py -i 37 -s 90 110 1 -g 20 60 2 (curvy boi but bad start angle, works with p=0.7)