CrowdWalk Gym provides a simple interface to instantiate Reinforcement Learning environments with CrowdWalk for crowd route guidance.
CrowdWalk is a multi-agent crowd movement simulator.
Please follow a documentation (https://github.com/crest-cassia/CrowdWalk).
git clone [email protected]:ryonsd/crowdwalk-gym.git
cd crowdwalk-gym
pip install -e .
$ cp -rp sample/* <path-to-your-CrowdWalk-dir>/crowdwalk/sample/ |
 |
|---|---|
| Two routes | Real map (moji port) |
When you try learning on "moji" environment, I recommend to use "moji_small" from the perspective of calculation time.
"moji" represents a real map and a generation of pedestrians, but it takes a long time to calculate a congestion degree from all pedestrian locations.
"moji_small" is a low-fidelity version of "moji", and the routes width and number of pedestrians are 1/10.
Its reduction makes the calculation time is shorter, 1 episode is 1 min.
If you test your algorithm tentatively, using "two_routes" is best because 1 episode is 12 seconds.
python dqn.py --simulator_dir <path-to-your-CrowdWalk-dir> --env_name two_routes --gui
The result of learning route guidance with DQN that minimizes congestion degree in "two_routes" and "moji" environment.
 |
 |
|---|---|
| Two routes | Real map (moji port) |