Tulsa STEM Partnership Launches Afterschool Robotics Clubs
On April 15, 2026, Memorial Middle and High School robotics teams secured first place in the VEX Robotics Regional Championship held at the Tulsa Expo Square, defeating 47 competing squads across Oklahoma and northwest Arkansas. Their victory wasn’t just a trophy moment—it represented the culmination of an 18-month STEM pipeline forged through Opp Project Tulsa Innovation Labs and PartnerTulsa, where students programmed autonomous routines using NVIDIA Jetson Orin Nano modules, tuned PID controllers on ROS 2 Humble Hawksbill, and integrated real-time computer vision via OpenCV 4.9 running on Ubuntu 22.04 LTS. This isn’t a feel-good extracurricular story. it’s a forward-deployed signal of how edge AI literacy is being baked into the next generation of systems thinkers—long before they hit college or a corporate helpdesk.
The Tech TL;DR:
- Student teams achieved 98.7% autonomous task success rate using Jetson Orin Nano (40 TOPS, 15W TDP) running custom YOLOv8n models for object detection.
- ROS 2 DDS middleware enabled sub-50ms inter-node latency between sensor fusion and motor control loops on Ubuntu 22.04 LTS.
- Funding and mentorship flowed through PartnerTulsa’s STEM grant program, backed by a $1.2M allocation from the Oklahoma Innovation Expansion Act (2024).
The real story lies in the technical stack these students wielded—hardware and software choices that mirror what’s running in industrial automation cells and autonomous warehouse bots today. Each team fielded two robots: one driver-controlled, one fully autonomous. The autonomous units relied on Jetson Orin Nano SOMs delivering 40 TOPS of AI performance at just 15 watts, a critical efficiency metric when battling for runtime in a six-minute match. Benchmarks from NVIDIA’s own Jetson Linux 36.2 documentation indicate the Orin Nano outperforming its Xavier NX predecessor by 80% in AI inference throughput while maintaining identical power envelopes—a detail not lost on the mentors guiding these teams. “We chose the Orin Nano specifically as its power-to-performance ratio lets us run concurrent perception and planning without thermal throttling,” said one lead mentor, a senior embedded systems engineer at a Tulsa-based aerospace supplier who requested anonymity. “In competition, if your vision pipeline drops frames because the SoC is overheating, you lose. Period.” This mirrors enterprise concerns where edge AI devices in manufacturing or logistics must sustain peak inference under load—failures here aren’t just lost points; they’re scrapped production lines or misrouted payloads.
ROS 2 DDS and the Latency Arms Race
Beneath the autonomous layer ran ROS 2 Humble Hawksbill, selected for its deterministic DDS (Data Distribution Service) middleware and improved real-time guarantees over ROS 1. Teams published sensor data (LiDAR, IMU, camera feeds) at 30Hz over UDP multicast, achieving measured end-to-end latency of 42ms from image capture to motor command publication—verified using ROS 2 tracing tools and hardware timestamps on the Jetson’s ARM Cortex-A78AE cores. This is significant: in autonomous mobile robot (AMR) applications, sub-50ms loop latency is often the threshold for stable dynamic obstacle avoidance. A 2023 IEEE Transactions on Robotics study found that latency exceeding 60ms increased collision risk by 300% in cluttered environments. The students didn’t just hit that mark—they beat it, tuning DDS QoS policies for reliability and disabling unnecessary debug topics to reclaim bandwidth. “We treated the robot like a micro datacenter,” explained a PartnerTulsa STEM coordinator during a post-match debrief. “Every cycle counts. You don’t acquire to reboot mid-match.”
Funding Lines and the Open Source Pipeline
Transparency matters. The robotics initiative isn’t volunteer-driven hobbyism—it’s a structured workforce development program. Funding originated from PartnerTulsa’s STEM Equity Fund, which received a $1.2M direct allocation under Section 103 of the Oklahoma Innovation Expansion Act (OIEA) of 2024, signed into law by Governor Stitt. This money covers hardware kits (Jetson Orin Nano developer boards at $599 each), software licenses (none—everything used is open source), and mentor stipends. Crucially, the software stack is entirely community-maintained: ROS 2 under Apache 2.0, OpenCV under BSD 3-clause, and Ubuntu LTS under Canonical’s standard licensing. No vendor lock-in. No black boxes. As one maintainer of the ROS 2 navigation stack noted in a public GitHub discussion thread last November, “The beauty of Humble is that a high school team in Tulsa can run the same nav2 code as a warehouse robot in Rotterdam—assuming they’ve tuned their PID gains.” That portability is the point. It mirrors how enterprise teams evaluate software: not by flashy demos, but by reproducibility, community health, and zero-cost scalability.
Directory Bridge: From Classroom to Cyber-Physical Security
This isn’t just about building better robots—it’s about building better defenders of cyber-physical systems. When students learn to harden ROS 2 nodes against rogue topics or validate input from camera feeds to prevent model poisoning, they’re learning fundamentals applicable to securing autonomous vehicles, industrial controllers, or drone swarms. And that’s where the directory comes in. For local manufacturers looking to deploy similar edge AI systems but lacking in-house expertise, firms like Tulsa-based MSPs specializing in OT/IT convergence can provide architecture reviews and deployment playbooks. Similarly, organizations adopting autonomous logistics need cybersecurity auditors familiar with ROS 2 threat models to audit DDS configurations and containerized workloads—especially as CISA warns of increasing targeting of ROS-based systems in critical infrastructure. Finally, schools or startups prototyping AI edge devices should consult embedded software agencies with Jetson ecosystem experience to avoid common pitfalls like thermal mismanagement or insecure default DDS permissions.
The implementation mandate: here’s how you’d verify DDS topic latency in a ROS 2 Humble environment—exactly what the students used to tune their perception pipeline:
# Measure latency between image capture and motor command ros2 run ros2_tracing latency_measurement --input-topic /camera/image_raw --output-topic /cmd_vel --history-depth 10 --period-ms 100 This command, part of the ros2_tracing package, publishes histogram data showing min, imply, and max latency—critical for proving real-time compliance. Teams ran this during practice iterations, adjusting DDS QoS and CPU affinity until the 95th percentile latency stayed under 50ms. It’s the kind of observable, measurable practice that separates engineering from tinkering.
The editorial kicker: as edge AI proliferates—from factory floors to farm fields—the ability to design, deploy, and defend these systems won’t live exclusively in PhD labs or Silicon Valley startups. It’s already being prototyped in afterschool programs where the biggest constraint isn’t funding or talent, but access to mentorship and real-world hardware. The Memorial teams didn’t win because they had the smartest kids; they won because they had the most coherent stack, the tightest feedback loop, and the discipline to treat a robot like a production system. That’s the standard we should be holding enterprise deployments to—and the directory exists to help you meet it.