Volume-DROID: All You Need is a Camera

---

Implementation of our paper: Volume-DROID

Authors: Peter Stratton (pstratt@umich.edu), Nibarkavi Naresh Babu Amutha (nibah@umich.edu), Ashwin Saxena (ashwinsa@umich.edu), Emaad Gerami (egerami@umich.edu), Sandilya Garimella (garimell@umich.edu)

Overview

Volume-DROID is a novel SLAM architecture created by combining the recent works: DROID-SLAM and NeuralBKI. Volume-DROID takes camera images (monocular or stereo) or frames from video as input and outputs online, 3D semantic mapping of the environment via combination of DROID-SLAM, point cloud registration, off-the-shelf semantic segmentation and ConvBKI. The novelty of our method lies in the fusion of DROID-SLAM and ConvBKI by the introduction of point cloud generation from RGB-Depth frames and optimized camera poses. By having only camera images or a stereo video as input, we achieved functional real-time online 3D semantic mapping.

All of our code is original, adapted from the NeuralBKI codebase, or adapted from the DROID-SLAM codebase.

NeuralBKI code adapted from: https://github.com/UMich-CURLY/NeuralBKI
DROID-SLAM code adapted from: https://github.com/princeton-vl/DROID-SLAM

Installation

Mandatory Installation

To install Volume-DROID, first clone the repo:

git clone https://github.com/peterstratton/Volume-DROID.git

Next, we need to install the eigen and and lietorch repositories. To do so, run the following commands from inside the Volume-DROID directory:

cd DROID-SLAM/thirdparty/
git clone https://gitlab.com/libeigen/eigen.git
git clone https://github.com/princeton-vl/lietorch.git

The last step of the mandatory installation is to download the sample of the neighborhood dataset from this link: https://cmu.app.box.com/s/5trtb7f3ogjao33lgk6xu6t9y9nu79wg. Then, from the Volume-DROID directory, create the directory using the following command:

mkdir datasets_withaccess

Last, unzip the downloaded neighborhood data sample, unzip it, and move it into the datasets_withaccess folder.

Optional Docker Setup

For our experiments, all of our code was run inside Docker containers. To install Docker on an Ubuntu machine, follow the instructions in this link: https://docs.docker.com/engine/install/ubuntu/#set-up-the-repository.

After Docker is installed, we need to build a Docker image with both pytorch and ros installed. We adapted a dockerfile given to us by a classmate for this purpose. To build the image, run the following command:

docker build -t torch_ros - < container.Dockerfile

Now that the image has been built. We can run the image to launch a container using the following commmand:

docker run --gpus all -it --rm --net=ros --env="DISPLAY=novnc:0.0" --env="ROS_MASTER_URI=http://roscore:11311" --name pstratt_volumedroid_torch_ros -v ~/Volume-DROID:/opt/Volume-DROID/ -v ~/Volume-DROID/services.sh:/opt/Volume-DROID/services.sh --ipc=host torch_ros /opt/Volume-DROID/services.sh --shm-size 16G

After running the above command, you should be in the /workspace directory inside the launched container. Run the following commands to move into the Volume-DROID directory:

cd ../opt/Volume-DROID/

Continued Mandatory Installation

Continuing from the Volume-DROID directory on either a local machine or launched docker continainer, we now need to create the conda environment and activate the environment:

conda env create -f environment.yaml
source /opt/conda/etc/profile.d/conda.sh
conda activate Volume-DROID

Next, we need to compile the third party libraries by running:

cd  DROID-SLAM
python setup.py install

Run the Code

To run the code, run the following command:

./tools/validate_tartanair.sh  --stereo