Third Place Solution in Spring 2026 Simulation Racing Series
Written by Mark Menaker, Aryan Mukherjee, and Radin Khosraviani, on behalf of their team from University High School (Irvine).
Github: https://github.com/MightyMark3/ROAR_1_S26
Table of Contents
Introduction and History
The 3rd place Spring 2026 solution is built off the code from our 1st place Fall 2025 solution. It uses an evolution of the throttle control logic that was originally introduced with the 2nd place Summer 2023 solution, as well as the steering controller from the Spring 2025 first place solution.
Main Adjustments
In order to traverse the track with more smooth steering adjustments, the steering controller was replaced by one used by George Zeng in Spring 2025, This controller steers toward a waypoint that is a weighted average of the points ahead, preventing abrupt steering transitions.
Our winning solution in Fall 2025 centered around left-foot braking. While braking on the way into a turn, we gradually increased throttle before lifting off the brakes, allowing for a quicker response when accelerating on the way out of the turn. This time around, we made the throttle buildup during braking periods more aggressive, improving lap times.
Another new idea we implemented was trail braking. Real trail braking involves gradually decreasing pressure on the brakes while turning, in order to keep braking during the turn while still keeping control, allowing a driver to start braking later. Our version of trail braking, however, involved progressively decreasing the braking value over the last 3 ticks. This allowed us to fine-tune braking, preventing the car from shedding too much speed and unnecessarily losing momentum before a turn.
Trial and Error
There are some ideas that we tested that only minimally affected the time to completion, so we would like to discuss them in this section so that future groups could potentially use them as a starting point to make further improvements.
For one, we noticed that on some sections of the track, the sensitivity of the steering looked as if there were non-linearities in the code and weren’t as smooth as others. To fix this, we went into the code corresponding to those sections in submission.py and tried shifting around the value of the steer multiplier by 0.01-0.02 values. Ultimately, the car still ran in a nearly identical fashion after making these changes so it is difficult to determine how effective they were, but the steering appeared to smoothen out slightly at the sections in question. At this point, the values are likely quite optimized, but it is possible that further modifications to that section could improve the time of the car.
Additionally, we went into the throttle controller part of the code and attempted to change the speed_up_threshold and throttle_decrease_multiple values, but this ultimately resulted in the car gaining time or crashing because they were already set and too many other parts of the code relied on those values to allow for effective tuning.
Finally, we tried making the car follow a path based on a smooth polynomial curve fit between waypoints k and k+3 on sections where the map waypoints split sharply (as determined by plotting them on Desmos) rather than having it follow a rigid line through the waypoints. This can be shown by the images below (yellow is the preferred solution, red is how the car operated currently in certain sections, as approximated by our sketch):
However, this proved to be difficult to implement and caused the car’s acceleration to slow down in the following section, so we decided to leave this idea for the next group of ROAR students at our school to experiment with.
Finally, it is worth noting that with the growing popularity of new artificial intelligence techniques to solve optimization problems, we considered the idea of deploying a gradient descent algorithm that would be trained on several simple predictor variables of time to complete the track. We could then use the model to optimize the features in question and determine the best values for achieving a fast completion time. However, we realized that this model would either be extremely weak or extremely complex depending on how many factors we accounted for, so we decided not to implement this solution.
Conclusion
A big thank you to Dr. Allen Yang, Mr. Huo Chao Kuan, and your team of experts for all the effort put into running the competition! As seniors who have competed since sophomore year, we enjoyed the opportunity to explore ROAR at our school. Another big Thank You goes to Mr. Shulman, who advises the Autonomous Car Racing Club at University High School! Finally, we would like to thank our other teammates: Pirouz Ruppert and Jocelyn Zhang.