NYU RoboMaster: The Ultimate Cross Between Gaming and Robots
Wait… what am I missing out on?
Let’s get you up to speed. You’re probably familiar with battlebots, an American television series in which each competitor designs a remote-controlled armored robot to compete in 1v1 battles with the goal of physically destroying or disabling an opponent’s robot. Now, imagine the same competition, but instead of creating robots with hammer-like arms to inflict physical damage, teams must tactfully build robots that use object-detection models to autonomously shoot small projectiles at enemy robots. Oh, and multiply the number of robots by three, or even up to nine. This is the reality not just for NYU’s RoboMaster team, UltraViolet, but all teams that partake in RoboMaster.
Alright, but what even is RoboMaster?
To preface, RoboMaster is one of the largest, annual student robotics competitions in the world, garnering over 2 million viewers each year. Established in 2013 and founded by DJI, RoboMaster initially only held competitions in China, but has now gained an international presence. University teams from across the world build several state-of-the-art robots to compete against dozens of other teams’ robots to take out an opponent’s base.
The RoboMaster University Series consists of two different kinds of RoboMaster competitions: RoboMaster University League (RMUL) and RoboMaster University Championship (RMUC). At the RMUL competition, there are 3 vs 3 matches. The 3 vs 3 matches are set up so that the Hero, Sentry and Standard (Infantry) robots (shown above) designed by one team battle the same 3 robots designed by another, all in one arena. Yes, it’s as chaotic as it sounds. There are also Standard matches, in which a Standard robot from each university team battles one another.
Sounds interesting, but what’s so special about RoboMaster at NYU?
UltraViolet, currently in its third year of operation, has 64 active members across 7 subteams. That’s a huge team, and yet, even building 3 robots with so much support is a challenge. Despite how young the team is, they competed at last year’s RMUL Lone Star competition and won 4th place overall. I got the chance to interview the Team Captain, Amy Zhao and Project Manager, Shaur Kumar to ask them about their experience as leads for the NYU RoboMaster team.
Shaur said that his interest in RoboMaster draws heavily from his FIRST Robotics background. “In 2021, I found out we had this thing called RoboMasters and I joined immediately… It looks just like FIRST Robotics and I want to keep that passion alive.” When Amy first heard about RoboMaster, building the robots was difficult due to the pandemic, but by no means did that dissolve her passion. “It’s a robotics competition, so why not just join it?” she says. For Amy and Shaur, robotics has been artfully woven into their lives.
Tell me more about this competition they’re working towards…
Of course, any conversation about RoboMaster lends itself to the competition, and Shaur and Amy had a lot to share about the team’s most recent competition. One of the highlights Shaur mentioned was being able to control the robot during battles. “As a driver, that’s always the most high adrenaline — it’s like playing a video game competitively.” While there is a lot of thought that goes into building the robots, the strategy behind driving the robots at competition can be equally important.
Amy said her favorite part of the competition was meeting other teams. “Previously we never met them in person, we were just communicating online, [but] we finally met them in person, talked to them, and looked at their robots.” The RoboMaster community is spread across the world, and the competition is really the best place to get to know other robotics fanatics.
When it comes to being successful in the competition, the build of the robot certainly matters. Shaur talked about improvements made to the two Standard robots that were built for the 2022 competition. “We took them apart because we were like we could make this much better if we changed these couple of things, and those couple of changes became its own design.” As more members join and ideas are bounced off of each other, the robot starts to take the shape of those enhancements.
I mean it’s a cool competition, but what is any of this doing in a data science-themed publication?
While the RoboMaster spotlight tends to land on what’s tangible, or the physical robot itself, the complex computer vision processes happening under the hood are what I’m personally here for. Every robot shoots autonomously, meaning that the robot relies on machine learning models to detect another robot’s armor plate (see above) and aim its turret at the plate to shoot.
To gain a better understanding of the computer vision portion of RoboMaster, I spoke with Ammaar Firozi, the Computer Vision (CV) Subteam Lead for NYU’s RoboMaster team. Ammaar gave insight on how the team went about processing data and training an object detection model.
“We used transfer learning with a yolov5 model on a crowdsourced dataset, most of which we label ourselves.” Image annotations and all preprocessing is done using Roboflow, designed specifically for robotics computer vision. The team uses YOLOv5, an object detection algorithm that trains the model for detecting robot armor plates. “YOLOv5 is really cool since it’s really lightweight, ‘You Only Look Once’ (not to be confused with the phrase “You Only Live Once”), so it’s good for mobile deployment in the case of robots.” The team trains the model to detect armor plates so that in competition, their robot can detect enemy plates in realtime from a camera feed.
Wow, this isn’t all that simple for a somewhat-underground robotics competition…
You’re telling me! Computer vision has several layers, especially in the context of RoboMaster. “It’s very easy to build something that can detect plates, but how we actually track them over time is a lot more difficult. That is sort of the distinction between perception and development.” Plate recognition is a core component of RoboMaster, but that alone is not nearly enough to surpass other teams. Ammaar brought up other challenges the team is trying to solve, such as creating “a tracking system using extended kalman filters. Other teams ‘beyblade,’ (when the base of the robot moves in 360-degree motion while the turret stays in one position) so unless you have some way to perceive fast changes, it’s pretty much impossible to hit them.” It’s important yet troublesome to also take into account a competitor’s capabilities.
The CV subteam certainly finds themselves applying skills they learn in the classroom, but solutions aren’t available at the back of a textbook. “What you’re taught in classes is how all these tools work. You can learn how to approach problems like classification. But shifting from theory to a real-time implementation is surprisingly difficult,” Ammaar says. As the CS Lead at NYU RoboMaster, I do most of my learning on the fly and will look at dozens of outside resources when it comes to brainstorming ideas. However, it’s this sense of drive and purpose that all CV team members share that makes it possible for the CV team to overcome hurdles. Ammaar told us, “Leading development outside classes with so many people on the product and seeing actual progress was very satisfying.”
Despite the motivation, this year the CV team really has their work cut out for them. The Sentry robot was originally on a rail above the base, but for this year’s competition, RoboMaster has asked teams to build a fully-autonomous Sentry ground robot. Ammaar highlights the team’s future plans to account for the rules change. “We have to do some sort of robot localization using SLAM with fiducial markers to do high-level tracking.”
It seems like the computer vision part is the hardest to figure out.
You would assume so, but every aspect of RoboMaster proves to be challenging. Shaur harps on how one of the biggest issues isn’t even building the robots themselves, but obtaining the necessary resources in order to make improvements. “Funding is a constant challenge, especially when you’re trying to build upwards of 5–9 robots.” Lots of materials get shipped from China, which also creates an additional shipping cost. The team is always looking for support and donations in the form of materials or monetization.
While Amy, Shaur, and Amaar were entranced by the core robotics component of RoboMaster, Shaur emphasizes that by no means is a robotics background required in order to join RoboMaster. “For example, we have an EE team for electrical engineers, we also have CS stuff for CS people, we’ve got mech stuff for mechanical engineers, and if you’re into more materials engineering we’ve got that as well alongside how we need to make composites and cut things properly. If you’re more into the side of non-engineering things, we’ve got our operations team which is all kinds of graphic design, web development, UI/UX design…”
So I don’t have to be a robotics genius to be a part of the team?
Not at all! When people hear RoboMaster, they tend to think that teams are mainly just roboticists, but Shaur expressed that the truth about who is involved in RoboMaster is contrary to the stereotype. “Pretty much anything you could be interested in as a person, we’ve got a spot for it on the team.”
Wanna keep up with NYU RoboMaster’s robot development? You can follow them on Twitter, Instagram, Tik Tok, or even Facebook. If you want to learn more about NYU RoboMaster or are considering joining in the upcoming semester, click here for more information!
References:
Roboflow, Robomasters Image Dataset (2023). Retrieved March 23, 2023 from https://app.roboflow.com/nyu-robomasters/robomasters-nc6sw/4
DJI, RoboMaster (2021). Retrieved March 23, 2023 from https://www.robomaster.com/en-US
NYU RoboMaster Ultraviolet team, NYU RoboMaster (2023). Retrieved March 23, 2023 from https://www.nyurobomaster.com/
Bhogal, R. (Interviewer). (2022, November 1), Shaurya Kumar.
Bhogal, R. (Interviewer). (2022, November 1), Amy Zhao.
Bhogal, R. (Interviewer). (2022, November 3), Ammaar Firozi.
