this repo makes for training and testing a Crowd navigation or a map based navigation

you can run online reinforcement learning in a navigation task by stable-baselines3 and sb3-contrib!

we have added GRU+PPO for user (please see result in my past repo)

1. ${\textsf{\color{pink}The pink}}$ can not be seen by 2d scan, ${\textsf{\color{blue}The blue}}$ can be seen by 2d scan. (this is for CrowdNav testing or training)

2. ${\textsf{\color{pink}A pink}}$ line is a path which is made by Dijkstra algorithm, and ${\textsf{\color{pink}A pink}}$ triangle is a subgoal which is made by look-ahead planning, ${\textsf{\color{Yellow}A yellow}}$ triangle is a goal.

3. apply map (with walls)

4. only walls (on the other way, only map is applied)

os: ubuntu20.04 , python: 3.8.x

mkdir catkin_ws && cd catkin_ws && mkdir src && cd src

git clone https://github.com/CAI23sbP/CrowdNav_Cai.git && cd CrowdNav_Cai && pip3 install -e . && cd .. 

git clone https://github.com/CAI23sbP/pymap2d.git && cd pymap2d && pip3 install -e . && cd ..

git clone https://github.com/danieldugas/Python-RVO2.git && cd Python-RVO2 && python3 setup.py build && python3 setup.py install && cd ..

git clone https://github.com/CAI23sbP/range_libc.git && cd range_libc/pywrapper && pip3 install -e. && cd ../..

see config file in crowd_nav/configs/*.yaml

train.py --model_name default_model --config_name base_config --device cuda

if you want to test example model, see detail : ### Example weight ###

test.py --n_eval 100 --weight_path Your_model_name --render True --config_name base_config

Here is testing rendering (from example weight)

see detail example_scan_sim.py (path: crowd_sim/envs/)

see detail example.py (path: crowd_sim/envs/policy/network_policies/)

if you want to use libraries which are caled imitation , stable-baslines3 and sb3-contrib, You must make the extractor. see detail example_extractor.py (path: drl_utils/algorithms/extractors/)

see detail base_config.yaml (path: crowd_nav/configs/)

see detail train.py

1. create new folder : model_weight / example

2. download model weight: Link

3. move weight to model_weight

4. test by under code

test.py --model_path example/Last.pt

[1] CrowdNav

[2] NavRep

[3] CrowdNav_DSRNN

[4] Pas_CrowdNav

snp0783@naver.com