Torc’s Innovative Advances in Autonomous Software Development
Torc’s autonomous software system incorporates elements of machine learning and artificial intelligence. The Torc Machine Learning Frameworks team is responsible for building a software stack that learns from data collected during the on-road testing of its truck fleet. This engineering team manages the automated training of machine learning models and oversees their testing and deployment to the embedded hardware in the trucks.
Nicolas Jourdan, Engineering Manager of the ML Frameworks team, stated, “Our aim is to facilitate rapid iterations of our autonomous software ML stack and enhance our training and deployment processes. This effort is essential for advancing the development of safe and reliable autonomous trucking technology.”
Pioneering New Techniques
The team’s work revolves around two initiatives: the Joint Training Framework (JTF) and the Joint Deployment Framework (JDF). The JTF redefines the training process for machine learning models, while the JDF innovates the deployment of these models in the autonomous Freightliner Cascadia trucks.
Recently, the team achieved a key milestone by conducting automated optimization and deployment tests using Hardware-in-the-Loop (HIL) benches. This new method allows the teams to perform deployment tests on hardware that mirrors the embedded systems in the trucks, instead of waiting for trucks to be available for each test. This system is intricately integrated within the team’s cloud workflows.
This advancement permits Torc to evaluate machine learning models in a simulated production environment more effectively and at a greater scale than previously possible.
Advancing Level 4 Autonomy
The efforts of the ML Frameworks team are vital for achieving Level 4 autonomous trucking capabilities on U.S. public roads. “Our frameworks and standards are the foundation that will facilitate swift software product releases,” emphasized Jourdan. “In the dynamic field of autonomous vehicle development, our ability to quickly iterate and deploy securely will distinguish Torc from its competitors.”
A Vision for Transformative Change
Fiete Botschen, Torc’s lead for the Machine Learning Training and Release Factory, emphasized the transformative impact of machine learning: “At Torc, we are not just developing autonomous vehicles. We are fostering a data-driven ecosystem that enables us to enhance our trucking software stack by utilizing the data collected by our trucks. This capability is fundamental for expanding our logistics network. Once our production trucks are on the road, we will be able to scale our operations rapidly.”
.
