# Hybrid strategy

**Repository Path**: bupt_htl/hybrid-strategy

## Basic Information

- **Project Name**: Hybrid strategy
- **Description**: No description available
- **Primary Language**: Python
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2023-03-18
- **Last Updated**: 2023-06-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Hybrid Strategy-Based Multi-Agent Decision Model

## Requirements
```
seaborn>=0.11.1
easydict>=1.9
matplotlib>=3.3.3
numpy>=1.19.4
scipy>=1.4.1
tensorboardX>=2.1
torch==1.7.1
torchvision==0.8.2
scikit-image>=0.14.0
scikit-learn>=0.19.1
hashids==1.3.1
torchsummaryX==1.3.0
```

## Dataset generation
Step 1: generate maps.
```
python generateScenarios/GenerateMap.py --random_map --gen_CasePool --gen_map_type random --chosen_solver ECBS --map_width 100 --map_density 0.1  --map_complexity 0.002 --num_dataset 5  --path_save /jinpeng/GNN/SCIMap
```
Step 2: Use expert algorithms to generate expert demonstration data for each map.
```
python generateScenarios/CasesSolver.py --loadmap_TYPE random --random_map --gen_CasePool --chosen_solver ECBS --map_width 20 --map_density 0.1  --map_complexity 0.005 --num_agents 10 --num_dataset 250 --num_caseSetup_pEnv 50 --path_save /jinpeng/GNN/SCIMap/solution --path_loadSourceMap /jinpeng/GNN/SCIMap/map100x100_density_p1
```
Step 3: Convert expert demonstration data from yaml format to mat format.
```
python generateScenarios/DataGenerateYaml.py --num_agents 10 --map_w 20 --map_density 0.1 --div_train 3 --div_valid 1 --div_test 2 --div_train_IDMap 0 --div_test_IDMap 1 --div_valid_IDMap 2 --solCases_dir /jinpeng/GNN/SCIMap/solution/ --dir_SaveData /jinpeng/GNN/SCIMap/ --guidance Project_G
```

Note that the ecbs file under the generateScenarios needs to be called at this stage. If your operating system cannot recognize the file, please follow https://github.com/whoenig/libMultiRobotPlanning regenerate ecbs file.


## Train and test
Model train
```
python main.py configs/Configs.json --mode train --best_epoch --map_w 20 --nGraphFilterTaps 3 --num_agents 10 --trained_num_agents 10 --batch_numAgent --extra_policy one
```
Model test
```
python main.py configs/Configs_Generalization.json --mode test --best_epoch --log_time_trained 1662818029  --nGraphFilterTaps 3  --trained_num_agents 10 --trained_map_w 20   --commR 7  --map_w 28 --num_agents  20 --num_testset 1000 --GSO_mode dist_GSO  --action_select exp_multinorm  --guidance Project_G --CNN_mode  Default --batch_numAgent --tb_ExpName GNN_Resnet_3Block_distGSO_baseline_128
```