Trucks and Drones (RL-TSP-VRP-D)

Using reinforcement learning to solve the travelling salesman and vehicle routing problem with drones (or robots).

Installation

• Install Trucks and Drones with pip:

pip install trucks-and-drones

• Install with git:

git clone https://github.com/maik97/trucks-and-drones.git
cd trucks-and-drones
python setup.py install

Dependencies

• gym >= 0.17.3
• (wacky-rl)

Documentation

https://trucks-and-drones.rtfd.io/

Example

from stable_baselines3 import PPO
from trucks_and_drones import BuildEnvironment

env = BuildEnvironment('test')

env.trucks(1)
env.depots(1)
env.customers(10)

env.observations(contin_inputs = [['v_coord'], ['c_coord'], ['d_coord'], ['demand']])
env.actions(discrete_outputs = ['nodes'])

env.compile()
env = env.build()

# Example: Proximal Policy Optimization using `stable-baselines3`
from stable_baselines3 import PPO
model = PPO("MlpPolicy", env, verbose=1)
model.learn(total_timesteps=200_000) #  takes about 10 minutes with cpu

# Alternative Example: Proximal Policy Optimization using `wacky-rl`
'''
from trucks_and_drones.wacky_ppo import WackyPPO
from wacky_rl.trainer import Trainer
from wacky_rl.logger import StatusPrinter

agent = WackyPPO(env, logger=StatusPrinter('WackyPPO'))
trainer = Trainer(env, agent)
trainer.n_step_train(200_000, train_on_test=False, render_env=False, render_test=False) #  takes about 25 minutes with cpu
trainer.agent.save_model()
'''

# Testing:

obs = env.reset()
sum_rewards = 0

for i in range(100):

    #action = agent.act(obs, act_argmax=True) #  for `wacky-rl` agents
    action, _states = model.predict(obs, deterministic=True) #  for `stable-baselines3` agents
    obs, reward, done, info = env.step(action)
    sum_rewards += reward

    env.render(slow_down_pls=True)
    if done:
        print('sum_rewards:', sum_rewards)
        sum_rewards = 0
        obs = env.reset()

env.close()

Citing

If you use trucks-and-drones in your research, you can cite it as follows:

@misc{schürmann2021trucks-and-drones,
    author = {Maik Schürmann},
    title = {trucks-and-drones},
    year = {2021},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/maik97/trucks-and-drones}},
}