The code thatdrives it.
Hardware wins matches. Software wins championships. Our codebase fuses computer vision, real-time SLAM, sensor fusion, and a custom architecture built for consistency, redundancy, and ease-of-use.
How we write code.
Consistency
Failing 5% of the time is failing all the time. Robust autonomous, adaptive control systems, sensor-based actions.
Optimal
Spend the least amount of effort for the most amount of points. Reliable subsystems, minimal wasted resources.
Redundancy
Better to have it twice than to need it once. Back-up sensors paired with mechanical and software solutions.
Ease-of-use
If your driver needs a manual, your code needs a redesign. Auto-advancing states, teleop sensor assistance.
Three layers. One unified system.
PHYSICAL LAYER
Hardware abstraction
Custom controls for every actuator — applicable to every season's robot. PIDs tunable in seconds.
LOGICAL LAYER
Subsystem brains
Deposit · Turret · Intake · Transfer · Vector. Each subsystem is independently testable and swappable.
COORDINATOR
The conductor
Sequences subsystems together for complex autonomous routines and assisted teleop actions.
The stack that wins matches.
Nine systems working together. Each one solving a specific problem we hit on the field, each one built from scratch by team members.
Real-Time SLAM
Simultaneous Localization & Mapping
Combined vision, depth sensing, and odometry to continuously map the field and localize the robot in real time. Allows us to adapt to alliance partners, update paths on the fly, and make informed autonomous decisions based on an accurate understanding of our surroundings.
Computer Vision
Limelight + HuskyLens AI
Limelight handles AprilTag detection for auto-aligning to the backboard and reading IDs to dynamically choose patterns. Ball detection identifies game pieces and feeds into the multi-zone sensor pipeline to compute SLAM updates.
Multi-Zone Depth Sensing
VL53L1X distance arrays
Multi-zone depth sensors detect approaching objects and measure their distance with sub-centimeter precision. Critical for collision avoidance during autonomous and for understanding the robot's micro-environment.
Pinpoint Localization
IMU + dead-wheel odometry
The Pinpoint computer fuses IMU and dead-wheel odometry into a single, precise localization system using global field coordinates rather than relative-to-robot. This is what powers our SLAM module.
Layered Architecture
Custom Physical Layer
Built our own layered architecture: a Physical Layer with custom controls applicable to every season, a Logical Layer on top, and a Coordinator that ties subsystems together. Enables fast prototyping and clean transfer between seasons.
SQLite Persistence
State across disconnects
Custom SQLite database persists transfer state and sensor data through robot disconnects between Auton and TeleOp. Static variables get reset on disconnect — the database doesn't.
Physics-Based Shooting
Adaptive auto-shoot
Physics-based shooting calculation with hood adjustments for lost velocity. We can shoot while moving — driver no longer needs to stop, eliminating opportunities for defenders to push us off line.
Agile Dev Pipeline
Branch-per-subsystem
Structured agile pipeline. Each coder works on their own branch per subsystem (Drivetrain, Intake, Deposit, Slides). The captain delegates, tests are written for each subsystem, and branches merge into main only after validation.
Bulk Reading + SRS Hub
Sub-10ms loop times
Switched from reading each sensor individually (slow loop times) to using SRS Hubs with bulk reading — reading every sensor in a single transaction. Dramatic improvement in loop frequency.
Five coders. One main branch.
With five people on our software team, multiple coders can't edit the robot at once. We follow an agile pipeline where the captain delegates subtasks to individual coders or pairs. Each one works on their own branch of the repository, builds the feature in isolation, writes tests for the subsystem, and merges into mainonly once it's validated.
What's next on the line.
Predictive trajectory optimization
Use SLAM history + game-state prediction to anticipate robot and alliance movements rather than reacting in real time.
Reinforcement learning agent
Train an RL model on past match data to improve endgame decisions and scoring strategies automatically — the robot gets smarter every match.