/learning_responsibility_allocation

Code base accompanying the ICRA 2023 paper titled "Learning Responsibility Allocations for Safe Human-Robot Interaction with Applications to Autonomous Driving"

Primary LanguagePython

Learning Responsibility Allocations

Learning Responsibility Allocations for Safe Human-Robot Interaction with Applications to Autonomous Driving

Ryan K. Cosner, Yuxiao Chen, Karen Leung, and Marco Pavone

This code base trains responsibility allocation functions $\gamma$ as described in our paper. Closed-loop simulation is currently unavailable as it relies on other people's unpublished code.

The pretrained model used to produce figures 2 and 3 and table 1 can be found here.

Installation

Install learning_responsibility_allocation

conda create -n lra python=3.8
conda activate lra
git clone git@github.com:rkcosner/learning_responsibility_allocation.git
cd learning_responsibility_allocation
pip install -e .

Download Static Scenes and unzip it in the learning_responsibility_allocation/tbsim/safety_funcs/ folder.

Install trajdata

cd ..
git clone git@github.com:NVlabs/trajdata.gittrajdata
cd trajdata
# replace requirements.txt with trajdata_requirements.txt included in tbsim
pip install -e .

nuScenes

  • Follow this link to the nuScenes dataset.
  • Register an account with nuscenes.
  • Download the US files for Full dataset (v1.0)>train_val 1 through 10 and the metadata, and the Map expansion pack v1.3
  • Organize the dataset directory as follows: 
    nuscenes/
│   maps/
|   |   basemap/
|   |   expansion/
|   |   prediction/
|   |   36092f0b03a857c6a3403e25b4b7aab3.png
|   |   37819e65e09e5547b8a3ceaefba56bb2.png
|   |   53992ee3023e5494b90c316c183be829.png
|   |   93406b464a165eaba6d9de76ca09f5da.png
|   samples/
|   sweeps/
│   v1.0-trainval/
learing_responsibility_allocation/
trajdata/

WandB

set up your weights and biases account using wandb login set your wandb api key export WANDB_APIKEY=<your api key>

2. Train responsibility model

For standard training with WandB enter the following in learning_responsibility_allocation/:

python3 scripts/train.py --dataset_path <path-to-nuscenes-data-directory> --config_name nusc_resp

If you do not want logging to WandB, then enter the same command with the --debug flag:

python3 scripts/train.py --dataset_path <path-to-nuscenes-data-directory> --config_name nusc_resp ---debug