MIDAS: Multi-agent Interaction-aware Decision-making with Adaptive Strategies for Urban Autonomous Navigation

Xiaoyi Chen, Pratik Chaudhari

GRASP Lab, University of Pennsylvania

ArXiV: https://arxiv.org/abs/2008.07081

Prepare the environment

Install Argoverse following the instructions here: https://github.com/argoai/argoverse-api

Install ffmpeg

Create collision and interaction sets

To create collision sets, change na on line 6 to be the number of agents in the environment, and change date on line 7 of road_interactions_environment/neighhood_v4_collision_set_gen.py. Then run python road_interactions_environment/neighhood_v4_collision_set_gen.py.
To create interaction sets, follow the steps in road_interactions_environment/neighhood_v4_interaction_set_creation.ipynb.

Train the model

In policy_network/neighborhood_v4_ddqn/train_tr5.py:

Change the filepaths on lines 237-243 to point to your generated collision sets, interaction set and evaluation set.
Change the environment and training hyperparameters from line 48 to 158 for your training purposes. The default values are for MIDAS. In order to run MLP, DeepSet, SocialAttention with the same hyperparameters, simply change the value of value_net on line 119 to vanilla, deep_set or social_attention.
Run python policy_network/neighborhood_v4_ddqn/train_tr5.py. Arguments:

    --date Training date
    --code ID of your experiment. Eg. c0-0
    --seed Experiment seed. Any integer between 0 and 65535.

Visualize an episode with a model checkpoint

In policy_network/neighborhood_v4_ddqn/visualize_episode.py:

Update the variables on lines 199-218 depending on the date, checkpoint ID and filepath, dataset filepath and the ids of the episodes that you want to visualize.
Run python policy_network/neighborhood_v4_ddqn/visualize_episode.py

Code References

Argoverse https://github.com/argoai/argoverse-api

Set Transformer https://github.com/juho-lee/set_transformer

License

This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-sa/4.0/ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.

grasp-lyrl / midas