Innovative, Interdisciplinary Collaboration Leads to Six-Time Win

Winning once can often be by chance, but to win six consecutive times in a row requires skills. Professor of computer science, Dr. CJ Chung, co-advisors Nick Paul and Joe DeRose, and several Lawrence Technological University students have won the Intelligent Ground Vehicle Competition (IGVC) this year for the sixth time in a row!

The competition was established in 1993 by the U.S. Army’s Combat Capabilities Development Command Ground Vehicle Systems Center. LTU’s Autonomous Campus Transport (ACTor) vehicle has been a continuous contestant at the IGVC competition since 2017 when the new “self-drive” challenge category was introduced.

LTU’s team consisted of team captain Justin Dombecki, who currently works at Discover and is an adjunct professor of computer science; Devson Butani, a 2019 LTU graduate in mechanical engineering who is currently attaining his Master of Science in Computer Science; Austin Ramsey, a recent 2023 Master of Science in Computer Science graduate; Ryan Kaddis, a fourth year in his accelerated five-year Bachelor and Master of Science in Computer Science program; and Adilur Choudhury, a 2023 Bachelor of Science in Computer Science graduate who is currently a software engineer at Cognex. The ACTor team met weekly on Tuesdays to set goals, discuss technologies to develop systems for the vehicle, collaborate on creative ideas, and divide tasks. Each member put in multiple hours a week of coding, testing, and documenting to accomplish their tasks for the ACTor.

Unlike motor sport competitions, IGVC is entirely autonomous, often requiring complicated and cutting-edge artificial intelligence algorithms, and is highly research focused. The competition has two categories: the auto-nav challenge and the self-drive challenge. The auto-nav challenge challenges a completely autonomous, unmanned ground robot to stay on an outdoor course while maneuvering around obstacles. The self-drive challenge challenges contestants to develop a smart, self-driving car able to autonomously change lanes, stop for stop signs, turn at intersections, maneuver around obstacles and pedestrians, self-park, and more.

To help win the competition this year, the ACTor team rewrote the programming for all sensors and computing components, implemented a new main computer with a GPU (a graphic processing unit) and an ergonomic swivel arm to hold the computer, programmed human and object detection using Yolo v8, developed an extension of Gazelle Sim (a lightweight 2D simulator), and researched lane-following alternatives through deep learning. Implementing and creating software for autonomous vehicles requires a system that can quickly adapt to new environments, behaviors, capability requirements, and physical hardware. This awareness led the LTU team to design an innovative layered and modular software architecture using Robot Operating Systems (ROS) and Lua script systems to specify vehicle behavior.

“Team members are smart, hardworking, responsible, and dedicated. [They] have professional programming skills. They volunteered their time and efforts even after the semester was over and the completion was held in early June,” Professor Chung states. Their efforts and countless hours of hard work did not go unnoticed; the team earned a $7,000 prize check this year from the IGVC organizer for their win.

Each team member was responsible for different aspects of developing the ACTor based on their skills and professional abilities. Dombecki was in charge of Lua engine (Core), Web interface, GPS and Waypoint control, camera sensing and image republishing, lane detection, lane centering and keeping, and the Drive By Wire system; Butani was in charge of mechanical and electrical systems, re-wiring, eStop, Yolo-based stop sign, and pedestrian detection; Ramsey was in charge of LiDAR-based obstacle detection and avoidance, and parking; Kaddis was in charge of turns, tire, and pothole detection and avoidance; and Choudhury was in charge of Yolo-based tire detection, pedestrian and pothole detection and avoidance, and the LED display system.

Given the complexity and size of the project, it is difficult for rookie teams to be successful; however, LTU’s team began working on the project earlier and towards the right direction due to the experienced co-advisors. Professor Nick Paul was the founding member from the inception of LTU’s involvement in 2017 when the team received a vehicle donation from Hyundai MOBIS. After getting a Master of Science in Computer Science from LTU in 2018, Paul became an adjunct professor and has been a co-advisor since. Professor Giuseppe “Joe” DeRose, an LTU adjunct professor of mechanical engineering, who is also an LTU computer science graduate student, was a key member in winning the 2021 and 2022 competition. This year he became a co-advisor to the team.

The competition challenges college students to design, build, and program autonomous vehicles that can be operated solely by autonomous control and are capable of navigating through a variety of outdoor barriers and conditions. The competition encourages participants to use interdisciplinary abilities (skills derived from engineering, computer science, artificial intelligence, and robotics) and enables them to create cutting-edge autonomous vehicle technologies. The competition also allows students to be research-oriented; the ACTor project spawned several research projects involving software engineering, machine vision, and deep learning. Research conducted by LTU students on the ACTor has been published.

Chung says, “Team members are volunteering their time even after the competition to write a journal paper and book chapter over the summer. They also successfully presented three posters for LTU’s research day in April.”

The ACTor team has worked diligently to develop an innovative autonomous vehicle that could one day be used on public roads. The team hopes to continue their success into next year’s competition.

by Nurzahan Rahman