Traversability_Application

System Information

• Operating System: Ubuntu 20.04
• ROS Version: Noetic
• GPU: NVIDIA 3090ti

Clone

git clone --recursive https://github.com/dongjineee/traversability_application.git

---

TODO

Gazebo Setup

For setting up the Gazebo simulation environment, clone the Husky repository:

https://github.com/dongjineee/husky

After cloning, follow the instructions on that page to run the Docker setup.

---

Image based

Fast Traversability Estimation for Wild Visual Navigation

Package RUN

##=========== wvn docker setting ===========##
cd traversability_application/wild_nav/wild_visual_navigation/docker
docker compose -f docker-compose-gui-nvidia.yaml build
docker compose -f docker-compose-gui-nvidia.yaml up -d
docker compose -f docker-compose-gui-nvidia.yaml exec wvn_nvidia /bin/bash
source first_run.sh

##=========== RUN SIMULATION ===========##
# In the sim_container
roslaunch husky_gazebo husky_lake.launch rviz:=wild_nav

# In the wvn_container
roslaunch wild_visual_navigation_jackal wild_visual_navigation.launch

Follow the Footprints: Self-supervised Traversability Estimation for Off-road Vehicle Navigation based on Geometric and Visual Cues

WayFASTER: A Self-Supervised Traversability Prediction for Increased Navigation Awareness

---

Geometric(Terrain structure) based

Learning Self-Supervised Traversability With Navigation Experiences of Mobile Robots: A Risk-Aware Self-Training Approach

Gaussian Process-Based Traversability Analysis for Terrain Mapless Navigation

---

Semantic based

These Maps are Made for Walking: Real-Time Terrain Property Estimation for Mobile Robots

GA-Nav: Efficient Terrain Segmentation for Robot Navigation in Unstructured Outdoor Environments

1. Dataset download

Please visit the official websites of the real-world dataset (RELLIS-3D) and the simulation dataset (NEGS-UGV) to download the files. For Data-set, we use ID annotations instead of color annotations. Please refer to the GANav Dataset Directory.

GANav Dataset Directory

GANav
├── data
│   ├── rellis
│   │   │── test.txt
│   │   │── train.txt
│   │   │── val.txt
│   │   │── annotation
│   │   │   ├── 00000 & 00001 & 00002 & 00003 & 00004 
│   │   │── image
│   │   │   ├── 00000 & 00001 & 00002 & 00003 & 00004 
│   ├── rugd
│   │   │── test_ours.txt
│   │   │── test.txt
│   │   │── train_ours.txt
│   │   │── train.txt
│   │   │── val_ours.txt
│   │   │── val.txt
│   │   │── RUGD_annotations
│   │   │   ├── creek & park-1/2/8 & trail-(1 & 3-7 & 9-15) & village
│   │   │── RUGD_frames-with-annotations
│   │   │   ├── creek & park-1/2/8 & trail-(1 & 3-7 & 9-15) & village
│   ├── goose
│   │   ├── goose_label_mapping.csv
│   │   ├── images
│   │   │   ├── train
│   │   │   └── val
│   │   ├── labels
│   │   │   ├── train
│   │   │   └── val
│   │   ├── LICENSE
│   │   ├── test.txt
│   │   ├── train.txt
│   │   │── val.txt
│   ├── lake
│   │   │── test.txt
│   │   │── train.txt
│   │   │── val.txt
│   │   │── annotation
│   │   └── image
├── configs
├── tools

2. Group semantic seg RUN

##=========== wvn docker setting ===========##
cd traversability_application/wild_nav/wild_visual_navigation/docker
docker compose -f docker-compose-gui-nvidia.yaml build
docker compose -f docker-compose-gui-nvidia.yaml up -d
docker compose -f docker-compose-gui-nvidia.yaml exec ga_nav /bin/bash

cd src/GANav-offroad/
pip install -e .

##=========== RUN Data processing ===========##
##for rellis-3d dataset
#run relable group4
python ./tools/convert_datasets/rellis_relabel4.py

#run relable group6
python ./tools/convert_datasets/rellis_relabel6.py

##=========== RUN Training ===========##
##for rellis-3d dataset(real_world)
python ./tools/train.py ./configs/ours/ganav_group6_rellis.py

##for lake dataset(simulation)
python ./tools/train.py ./configs/ours/ganav_group6_lake.py

##=========== RUN Eval ===========##
##for rellis-3d dataset
python ./tools/test.py ./trained_models/rellis_group6/ganav_rellis.py \
          ./work_dirs/ganav_group6_rellis/latest.pth --eval=mIoU

##for lake dataset
python ./tools/test.py ./trained_models/lake_group6/ganav_lake_6.py \
          ./work_dirs/ganav_group6_lake/latest.pth --eval=mIoU

##=========== RUN Visualize ===========##
python ./tools/visualize.py <img_dir> <config> <checkpoint>
##for rellis-3d dataset
python ./tools/visualize.py ./data/rellis/image/00000 ./configs/ours/ganav_group6_rellis.py ./work_dirs/ganav_group6_rellis/latest.pth

##for lake dataset
python ./tools/visualize.py ./data/lake/image ./configs/ours/ganav_group6_lake.py ./work_dirs/ganav_group6_lake/latest.pth

3. ROS_PKG RUN

# In the sim_container
roslaunch husky_gazebo husky_lake.launch rviz:=ga_nav

# In the ga_nav_container
roslaunch ga_nav ga_nav.launch

---

Geometric + Semantic based

Probabilistic Traversability Model for Risk-Aware Motion Planning in Off-Road Environments

Multitask Learning for Scalable and Dense Multilayer Bayesian Map Inference

---

Real World - Dataset

• RELLIS-3D : Data with Stereo Camera images, LiDAR pointclouds, GPS/IMU
• RUGD : Video dataset annotated with pixel-wise labels

Simulation World - Dataset

• NEGS-UGV : semantic & rgb images, seamantic & raw Lidar pointsclouds