MAPP demonstrations (training data), benchmark instances, and trained models are available below.

bash scripts/exp_scripts/data_gen_hetero.sh    # required time: half day by x40 multiprocessing, seed=100000

The data will be stored in /data/demonstrations/learn_hetero/21-30.

bash scripts/exp_scripts/learn_hetero.sh cuda:0      # required time: 2 hours, you can use cpu instead of cuda

The repo includes the trained model in /workspace/trained_model/aamas22-main.

bash scripts/exp_scripts/benchmark_gen_hetero.sh   # seed=46

The data will be stored in /data/benchmark.

All the results will be saved in /data/exp.

bash scripts/exp_scripts/eval_ctrm_large_learned_ind.sh    # required time: 1 day

The used trained model is ctrm_data/models/with_ind_k15.

bash scripts/exp_scripts/eval_random_large.sh

bash scripts/exp_scripts/eval_grid_large.sh

bash scripts/exp_scripts/eval_spars_large.sh

In the heterogeneous scenario, the method uses multiprocessing. Although this affects runtime, the method anyway results in a low success rate (hence excluded in the figures with quality metrics).

bash scripts/exp_scripts/eval_square_large.sh

bash scripts/exp_scripts/learn_hetero_wo_comm.sh         # without communication
bash scripts/exp_scripts/learn_hetero_wo_indicator.sh    # without indicator

The trained models are already included in ctrm_data/models.

bash scripts/exp_scripts/eval_ctrm_ablation_learned_ind.sh

1. Karaman, S., & Frazzoli, E. (2011). Sampling-based algorithms for optimal motion planning. The international journal of robotics research (IJRR)
2. Dobson, A., Krontiris, A., & Bekris, K. E. (2013). Sparse roadmap spanners. In Algorithmic Foundations of Robotics X.