Share this
Loadsmart has been a partner of the MIT Transportation Center since 2023. This partnership helps provide students with data and challenges related to real-world situations. Students are presented with challenges and are tasked with building machine learning models to solve them. Our team watches the presentations and provides the students with feedback to help their studies. This year, Loadsmart has proposed an ETA-based challenge that incorporates tracking data.
As part of the partnership, Loadsmart was invited by MIT Professor Elenna Dugundji to present NavTrac to a class of students beginning their studies in the computer vision field earlier this month. The session gave students a firsthand look at how NavTrac applies computer vision in real-world settings.
Our engineering team, including Rodrigo Senra (VP of Engineering), Gustavo Valiati (Senior Engineer II), Diogo Mafra (Staff Engineer), Hai Nguyen (Engineering Manager), and Aaron Thomas (Director of Engineering), hosted the presentation and were available to answer questions from the MIT students.
We reviewed how NavTrac uses computer vision to track every vehicle entering or leaving a yard. The system captures images from multiple cameras. It identifies key details like license plates, container and chassis numbers, and trailer IDs. Our devices process this data on-site using NVIDIA Jetson hardware. Then the system sends structured records to the cloud for NavTrac customers to review.
During the session, we explained how we train our models using more than 30,000 authentic images for object detection and over 300,000 samples for OCR. We include synthetic data to help models handle rare or inconsistent formats. We also shared how we monitor quality, with our team reviewing some events by hand to ensure accuracy and iterative improvements over time.
We also talked about our internal tools. One tool lets us play back multi-camera footage with detection overlays. Another lets our team review and correct model errors. These tools help us improve performance without slowing down deployments. Moreover, historical footage can be used to inspect damage in vehicles.
We thank Professor Dugundji and MIT for the invitation. It gave us a chance to show how AI works outside the lab in living, breathing use cases. Our team appreciates the opportunity to share what we’ve built and how we continue to improve as OCR capabilities evolve!
NavTrac supports on-the-ground logistics teams across the country every day. Our technology isn’t research demos or early-stage prototypes. NavTrac is a complete system that performs under pressure and delivers accurate results in real conditions.
Watch the full session here:
