RL Coverage Path Planning

Code implementation for the paper Learning Coverage Paths in Unknown Environments with Deep Reinforcement Learning, arXiv, 2023.

Lawn mowing	Exploration

WITHOUT total variation reward	WITH total variation reward

Install

• Install miniconda
• conda create -n rlm python=3.9
• conda activate rlm
• (CPU) pip install torch==1.13.1+cpu torchvision==0.14.1+cpu -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cpu
• (GPU) pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu116

Run

Train an RL agent for CPP in a simulated 2D environment (reduce --buffer_size if memory is limited):

• Mowing: python train.py --logdir my-experiment
• Exploration: python train.py --logdir my-experiment --exploration --local_tv_reward_scale 0.2 --no-overlap_observation --no-steering_limits_lin_vel

Check how the trained agent performs:

• Plot logs: python plot.py --load my-experiment
• Render: python eval.py --load my-experiment

Tests

Run tests:

• Run pytest from the root folder.