Edge-enhanced Attentions

An edge-enhanced attention model to capture the heterogeneous relationships among various nodes (e.g. pickup, delivery, charging) in drone delivery. Training with REINFORCE algorithm while in consideration of stochastic wind conditions.

Paper

For more details please see our paper:

Liu, R., Shin, H.S. and Tsourdos, A., 2023. Edge-Enhanced Attentions for Drone Delivery in Presence of Winds and Recharging Stations. Journal of Aerospace Information Systems, 20(4), pp.216-228. Publisher Link

Dependencies

• Python
• NumPy
• SciPy
• PyTorch
• tqdm
• tensorboard_logger
• Matplotlib (optional, only for plotting)

Quick Start

For training E-PDP instances with 20 nodes and using the proposed AM-E model:

python run.py --problem epdp --graph_size 20  --baseline rollout --run_name 'EPDP20_rollout' --model attention --attention_type withedge1

Usage

Training

For training E-PDP instances with 20 nodes and using the proposed AM-E model:

python run.py --problem epdp --graph_size 20  --baseline rollout --run_name 'EPDP20_rollout' --model attention --attention_type withedge1

Resume training

You can resume a run using a pre-trained model by using --resume:

python run.py --problem epdp --graph_size 20 --baseline critic --resume 'outputs/sepdp_20/epoch-65.pt'  --model GPN --n_epochs 100 

Validation

Generating data

To generate test data for all problems:

python generate_data.py --name test --problem epdp --graph_sizes 20 --seed 6666 --dataset_size 1000

Evaluation

To evaluate the performance, we run eval.py on test data:

python eval.py data/epdp/epdp20_test_seed6666.pkl --model 'outputs/epdp_20/epoch-99.pt' --model_type attention --decode_strategy greedy --eval_batch_size 1

Baselines

Baselines using Or-tools can be run as follows:

python epdp_baseline.py ortools10 'data/epdp/epdp20_test_seed6666.pkl' -f --battery_margin 0.2 --problem epdp