Algorithm package for MDP

Install the package in development mode

Ensure you have cloned and install imagerec package from this repo.

In the root folder containing setup.py file, run

pip install -r requirements.txt
pip install -e .

The flag -e is short for --editable, and . refers to the current working directory, so together, it means to install the current directory in editable mode. This will also install any dependencies declared with install_requires

Week 8 task

Note: Before running the main file, make sure to remove the images in "images" folder.

python -m mdpalgo

Week 9 task

Note: Before running the main file, make sure to remove the images in "images" folder.

python -m fastestalgo

Details to note

The arena:

• The map is a 20 by 20 grid of cells
• Each cell represents 10cm x 10cm

The robot:

• Modeled as a 3x3 on the grid
• Actual robot size is about 20cm x 21cm
• Turning radius:
  • Notes suggested about 25cm turning radius
  • In the simulator, the turning radius used is 3x3 which is 30cm by 30cm

The "obstacle" model:

• Physical size is identical to size on grid (1x1) (10cm x 10cm)
• Obstacle border given for astar path planning is about 20cm (2 cells away)

References

This repo is based upon a previous repo: https://github.com/CZ3004-Group-12/mdp-algorithm working on the same project in the past. Having to keep the developments private during CZ3004 course, we did not fork the repo, but we tried to keep the commit history intact.

We have contributed significantly, especially on the following aspects:

1. Fix the path planning algorithm: the A star algorithm in the previous repo did not immediately generate a sequence of feasible movements, but an idealised path for a holonomic robot, and converted to the actual path using a lot of hardcoding. Lots of bugs in the actual path, and the car sometimes crashed into the obstacle. We have completely revamped the A start path planning algorithm to immediately generate a sequence of feasible movements, with no collision in the simulator.

2. Develop our own approach for week 9 task. The algorithm side simply handles communication with RPi and calls the image recognition and leaves the planning to the RPi and STM.

3. Repackage the entire repo as two packages that are easy to install and deploy.

We also added a lot of unit tests and fixed numerous bugs in the process.