Kensuke Nakamura,
Somil Bansal
University of Southern California
This is the open-source repository for the paper "Online Update of Safety Assurances Using Confidence-Based Predictions".
Setup DeepReach. The instructions can be found the README here
Setup Trajectron++. The instructions can be found here.
You will need to add Trajectron++ to this directory. We use pretrained directory found in the eccv2020 branch Trajectron-plus-plus/experiments/nuScenes/models/int_ee_me/model_registrar-12.pt
In addition, you will need to replace the trajectron.py and mgcvae.py files with the ones provided in this repo. The location to place these are Trajectron-plus-plus/trajectron/model/FILENAME. These changes simply skip the final step in the Trajectron++ prediction to allow a direct output of the predicted control actions, rather than integrating them through the dynamics to get predicted occupancies.
Furthermore, we provide two scenarios based on the original NuScenes dataset. These can be found in the two pkl files, and should be added into a directory like so: Trajectron-plus-plus/experiments/nuScenes/processed/FILENAME.pkl
You will need to create the processed directory. This takes the original NuScenes scenarios and processes the data into something Trajectron++ can parse.
The directory experiment_scripts contains one script per experiment in the paper.
To monitor progress, the training code writes tensorboard summaries into a "summaries"" subdirectory in the logging_root.
To start training DeepReach for the extended unicycle model run:
CUDA_VISIBLE_DEVICES=0 python experiment_scripts/train_dubins4dForwardSetParam2set_scaled.py --experiment_name experiment_1 --minWith target --num_src_samples 12000 --pretrain --pretrain_iters 40000 --num_epochs 150000 --counter_end 110000 --periodic_boundary --adjust_relative_grads --diffModel --diffModel_mode 'mode2' --collisionR 0.17
This will regularly save checkpoints in the directory specified by the rootpath in the script, in a subdirectory "experiment_1".
We also provide a pretrained FRT, which is in logs/dubins4dParamFRS_pre40_src12_epo150_rad0017_2set_adjgrad_scaled_time
To run our examples, run
cd post_processing
python FRT_Query_final.py
This will spit out the numerical data given in our paper.
If you find our work useful in your research, please cite:
@misc{nakamura2022onlineupdate,
doi = {10.48550/ARXIV.2210.01199},
url = {https://arxiv.org/abs/2210.01199},
author = {Nakamura, Kensuke and Bansal, Somil},
keywords = {Robotics (cs.RO), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {Online Update of Safety Assurances Using Confidence-Based Predictions},
publisher = {arXiv},
year = {2022},
copyright = {Creative Commons Attribution 4.0 International}
}
If you have any questions, please feel free to email at k.nakamura@princeton.edu