by Samuel Joseph Mcmurray
This project was written in python utilizing docker to create a reproducable environment, included is a requirements.txt containing the versions used to produce the results. The requirements.txt was produced from the Dockerfile and would need to be run with it if the later versions fail to run.
This paper has been submitted for publication in IEEE Intelligent Transportation Systems with revisions and changes from this Thesis. This work looks at the effects of incorporating environmental infrastructure data into Transformer models for Multi-Class Trajectory Prediction. How these effects influence predictions in agents interactions with other agents and how agents interact with the static objects such as trees, signs, poles.
Caption for the example figure with the main results.
Trajectory Prediction (TP) plays a pivotal role across various domains, transforming navigation and interaction within complex environments. This study introduces the Semantic Aware Environ- mental Spatial-Temporal Graph Transformer (SAE-STAR) model, designed for multi-class TP in dy- namic urban settings. By leveraging deep learning and environmental infrastructure data, the model forecasts the movements of pedestrians, bicyclists, and diverse vehicle types. Research questions ex- plore enhancing prediction accuracy through infrastructure data, optimization of multi-class models, and forecasting agent-infrastructure interactions. Integrating graph-based and deep learning techniques aims to overcome existing TP model limitations, contributing to more accurate and reliable prediction systems. Empirical studies and real-world experiments provide insights into TP capabilities, limita- tions, and potential impacts on intelligent systems and decision-making processes. The study identifies improvements in prediction accuracy with environmental data integration, notably demonstrating the superior performance of the SAE-STAR model on the Valhallavägen dataset compared to Semantic Aware Spatial Temporal Graph Transformer (SA-STAR). Challenges include class imbalance effects, complexities in static feature incorporation, and hyperparameter tuning difficulties. Quantitative analy- sis shows an effective prediction of linear trajectories but challenges in complex scenarios corroborated qualitatively. Future work entails refining model architectures, extensive hyperparameter optimization, and enhancing data collection methodologies to improve TP model robustness in urban environments.
Briefly describe the software that was written to produce the results of this paper.
All source code used to generate the results and figures in the paper are in
the code folder.
The calculations and figure generation are all run inside
Jupyter notebooks.
The data used in this study are privately owned and maintained by Viscando AB and is not available for the replication and use.
Results generated by the code are saved in Results.
You can download a copy of all the files in this repository by cloning the git repository:
git clone https://github.com/pinga-lab/PAPER-REPO.git
A copy of the repository is also archived at insert DOI here
You'll need a working Python environment to run the code.
To set up your environment in anaconda is through the
Anaconda Python distribution which
provides the conda package manager.
Anaconda can be installed in your user directory and does not interfere with
the system Python installation.
The required dependencies are specified in the file environment.yml.
We use conda virtual environments to manage the project dependencies in
isolation.
Thus, you can install our dependencies without causing conflicts with your
setup (even with different Python versions).
Run the following command in the code repository folder (where environment.yaml
is located) to create a separate environment and install all required
dependencies in it:
conda env create -f environment.yaml
Before running any code you must activate the conda environment:
conda activate thesis
This will enable the environment for your current terminal session. Any subsequent commands will use software that is installed in the environment.
The virtual environment can be installed using pip if you have not installed a virtual environment before you will need to install the package from pip:
pip install virtualenv
Windows:
python -m venv venv_name
In CMD to activate the venv use:
venv_name\Scripts\activate.bat
Powershell to activate the venv use:
venv\Scripts\Activate.ps1
In linux or Mac:
virtualenv venv_name
source venv_name/bin/activate
This will enable the environment for your current terminal session. Any subsequent commands will use software that is installed in the environment. To install the python packages for the experiment you will need to be in the code directory where :
pip install -r requirements.txt
To set up your environment in anaconda is through the
Get started with Docker which
provides the Docker download links and documentation.
Using Docker offers numerous benefits by enabling consistent and efficient application deployment. It packages applications and their dependencies into lightweight, portable containers, ensuring they run reliably across different environments, from development to production. Docker simplifies collaboration by providing a standardized environment, reducing "it works on my machine" issues. It enhances scalability by allowing rapid scaling of services through container orchestration tools like Kubernetes. Additionally, Docker improves resource efficiency by sharing the host OS kernel, enabling faster start-up times and reduced overhead compared to traditional virtual machines. Its ease of integration with CI/CD pipelines accelerates development workflows, fostering agile software delivery.
The requirements.txt file is a dependency for the implementation of the environment for the code all that is need to run the docker environment is to have it in the folder docker will do the rest. To execute the docker container:
docker compose up -d
This will build the image and start the container the download of the image takes some time with many packages to download and install but it is extremely easy. Although 1 problem that I encountered is a problem with opencv to that follow these commands:
docker exec -it ai-gpu-container bash
pip uninstall $(pip list --format=freeze | grep opencv)
rm -rf /usr/local/lib/python3.10/dist-packages/cv2/
pip install opencv-python-headless
The three commands uninstalling the opencv package can also be executed in docker desktop by going to the container and going the exec tab.
To run the experiment in full you can execute the main.py file to execute the
code in sequentual order this takes a lot of time and is resource intensive and requires
the Viscando AB data although the experiment could be modified based to replicate with
another dataset.
Another way of exploring the code results is to execute the Jupyter notebooks individually. To do this, you must first start the notebook server by going into the repository top level and running:
jupyter notebook
If the method used is Docker the jupyter notebook is running and all you need to
do is to go to Hosted Jupyter Notebook
This will start the server and open your default web browser to the Jupyter
interface. In the page, go into the notebooks folder and select the
notebook that you wish to view/run.
The notebook is divided into cells (some have text while other have code).
Each cell can be executed using Shift + Enter.
Executing text cells does nothing and executing code cells runs the code
and produces it's output.
To execute the whole notebook, run all cells in order.
All source code is made available under a BSD 3-clause license. You can freely
use and modify the code, without warranty, so long as you provide attribution
to the authors. See LICENSE.md for the full license text.
The manuscript text is not open source. The authors reserve the rights to the article content, which is currently submitted for publication in the JOURNAL NAME.