Installation
Downloading Data
I download the following and stored at the YOUR_DATA_ROOT
Metadata for the mini split (v0.2) [US, Asia] 5.40 GB(5792841728 Bytes)
Metadata for the mini split (v0.1) [US, Asia] 5.39 GB(5792751616 Bytes)
nuPlan maps, required for mini and full dataset [US, Asia] 1.21 GB(1297742811 Bytes)
Env Setup
pip install -e .
pip install scipy==1.8.0
pip install pytorch-lightning==1.5.10
pip install einops
export NUPLAN_DATA_ROOT="<YOUR_DATA_ROOT>"
export NUPLAN_ROOT="<YOUR REPO/nuplan ROOT>"
export PATH=$HOME/.local/bin:$PATH
- Example for Environment Variable for reference.
NUPLAN_ROOT=~/MSU/course/autonomous_vehicle/nuplan-devkit_v2/nuplan
NUPLAN_DATA_ROOT=/data/data/autod
Running Experiments
The repo is structured as follows.
Run scripts
nuplan_devkit
├── ci
├── docs
├── nuplan
│ ├── common
│ ├── database
│ └── planning
└── tutorials - Where all the training scripts are.
The training scripts are in tutorials/.
Check out below for running attention experiments.
cd tutorials/
bash run_attention.sh
Network Architecture Change
The modification in the network architecture is in
nuplan_devkit
├── ci
├── docs
├── nuplan
│ ├── common
│ ├── database
│ └── planning - Where training codes are
└── tutorials
Some important files are:
nuplan/planning/script/config/common/model/ : contains yaml file for model arch. (attention configuartion is set here)
nuplan/planning/script/config/training/lightning/default_lightning.yaml : default config related to training.
nuplan/planning/training/modeling/models/rater_model.py : where rater_model forward is determined.
nuplan/planning/training/modeling/models/resnet.py : resnet redefined for attention.
nuplan/planning/training/modeling/models/attentions.py : attention modules.