V5 Override AI Roadmap

VEX V5RC 2026-27

Own the Override season with scouting first, driver assist second.

Ethan's best AI angle is a public, tournament-ready scouting and match-prediction app, followed by an on-robot strategy assistant that turns Toggle state, yellow Pins, and field position into clear driver calls.

Season plan Project rankings
Red Blue T T T T Midfield
8 per robot
56 Cups 63 Pins 9 Goals 4 Toggles

Game Intel

Override rewards fast information, then precise scoring.

Confirmed game shape

  • 12 ft by 12 ft field, two alliances, two teams per alliance.
  • 15-second autonomous period, then 1:45 driver control.
  • 56 Cups, 63 Pins, 9 Goals, 4 wall Toggles, 4 Loaders, and a contested Midfield.
  • Official manual link now shows version 0.1.2, released May 14, 2026.

Public scoring summary

  • Alliance-color Pin in a Goal: 5 points.
  • Owned yellow Pin in a Goal: 10 points, gated by the matching quadrant Toggle.
  • Robot ending in Midfield: 8 points.
  • Autonomous Bonus: 12 points. Autonomous Win Point remains a separate objective.

AI Leverage

The legal on-robot AI surface is narrow. The off-robot surface is huge.

1

Cup and Pin discrimination

Yellow Pins can be worth twice as much as alliance-color Pins. AI Vision Sensor classification plus color signatures should guide target priority.

2

Toggle-quadrant planning

The core strategic question is which Toggle unlocks the densest remaining yellow-Pin opportunity, then how quickly the robot can convert that state into points.

3

Autonomous path quality

LemLib-style odometry and closed-loop visual alignment beat open-loop routines once the team has stable mechanisms and field measurements.

4

Midfield endgame response

The last 20 seconds reward position reads and quick decisions. A driver assistant can turn clock, field state, and opponent tendency into one simple call.

5

55 W drivetrain cap

A power cap raises the value of smooth routing, lower-stall driving, and scoring-density plans over raw speed.

Legal Boundary

Build around the V5RC rule boundary.

Use on the robot

VEX AI Vision Sensor, GPS Sensor, Inertial Sensor, Rotation Sensor, Distance Sensor, and Optical Sensor.

Keep off the robot

Jetson, Raspberry Pi, phones, laptops, and custom coprocessors belong in scouting, dashboards, simulation, and planning for standard V5RC.

Code stack

PROS and C++ are the strongest Codex-friendly path because LemLib, OkapiLib, and other serious VEX libraries live there.

Project Menu

Six shippable builds, ranked by season value.

Pick 1, spine Scouting + Match-Prediction Web App Pull RobotEvents data, rate teams, rank alliance picks, and predict 2v2 matchups. High value / low risk

What Ethan builds

A phone-friendly tournament dashboard with event search, team profiles, Skills history, match history, manual scout notes, predicted match odds, and an alliance pick board that can be sorted in the last five minutes before selections.

Inputs

  • RobotEvents API: events, teams, rankings, matches, awards, and Skills.
  • Manual scouting: auton reliability, scoring style, defense, and endgame.
  • Derived features: win rate, schedule strength, AWP rate, and score spread.

First build

  • Week 1: event lookup, team pages, and data cache.
  • Week 2: basic Elo, matchup odds, and pick-board export.
  • Week 3: TrueSkill, manual tags, and tournament notes.

Definition of done

Before an event, the app lists every team Ethan might face, flags likely alliance partners, and produces a ranked pick list that the drive team actually uses.

Why it wins: this project does not depend on the robot being finished, so Ethan can ship it early and improve it every tournament.

Pick 2, complement Override Match-Strategy Auto-Pilot Turn Toggle state, yellow-Pin opportunity, odometry, and match clock into driver calls. High value / medium risk

What Ethan builds

A rule-based driver assistant that keeps a small field model and shows one next action on the V5 Brain: target yellow Pins, flip a Toggle, return to Loader, defend Midfield, or abandon a low-value cycle.

Inputs

  • AI Vision Sensor detections for Pins, Cups, Goals, and visual alignment.
  • Odometry from rotation sensors plus IMU heading.
  • Match clock, alliance color, Toggle ownership, and scouted opponent habits.

First build

  • Start with a simulator that accepts fake field state.
  • Implement a scoring heuristic: points per second plus travel penalty.
  • Move to V5 Brain display once the robot has stable odometry.

Risk to manage

Bad advice is worse than no advice during a match. Keep the first version conservative, visible, and easy for the driver to ignore.

Why it wins: drivers make repeated tactical choices under pressure. A clear nudge can turn field awareness into points without taking control away from the driver.

Layer 2 Autonomous Routine Generator Generate 15-second LemLib-style routes from starts, mechanisms, and field geometry. High value / medium risk

What Ethan builds

A laptop planner that turns a start tile, target order, and mechanism timing into waypoints, turn headings, intake commands, scoring commands, and an exported C++ routine for PROS.

Inputs

  • Field CAD measurements and a hand-checked coordinate map.
  • Robot constraints: max velocity, turn rate, intake time, and score time.
  • Opponent auton tendencies from the scouting app.

First build

  • Write a top-down field editor with draggable goals and objects.
  • Export Pure Pursuit waypoints and action markers.
  • Replay the route in a simple 2D sim before flashing the robot.

Definition of done

Ethan can generate three viable autonomous routines for different partners or sides, test them in sim, and bring the best one to the field for tuning.

Why it wins: most teams tune a small number of autonomous routes. A generator makes match-specific routine selection realistic.

Layer 2 Driver Practice Tracker + Auto-Critique Convert practice logs into measurable driver improvement and short coaching notes. High value / low risk

What Ethan builds

A telemetry logger and analysis dashboard that tracks driver inputs, motor temperatures, wattage, stopped time, quadrant occupancy, target misses, and scoring-cycle timing across practice runs.

Inputs

  • V5 Brain logs written to SD card or streamed over USB.
  • Practice match annotations: route, partner, defensive pressure, and result.
  • Optional AI Vision events: target acquired, target lost, and alignment time.

First build

  • Log joystick and motor metrics at 10 Hz.
  • Graph idle time, heat, wattage spikes, and cycle length.
  • Generate a three-bullet critique after each session.

Definition of done

After ten practice runs, Ethan can show where the driver improved, which route wastes time, and which mechanism problem is limiting score.

Why it wins: driver practice becomes visible. The team can train the exact behavior that costs points instead of relying on memory after a messy run.

Specialist Opponent-Behavior Classifier Label opponents from motion patterns, zone occupancy, and scouting priors. Medium value / medium risk

What Ethan builds

A classifier that identifies common opponent modes: defender, Loader cycler, Toggle camper, yellow-Pin hunter, or Midfield closer. The first version should be heuristics, then a small model once enough match logs exist.

Inputs

  • Manual scout labels from prior matches.
  • Zone occupancy and rough robot motion from vision or driver annotation.
  • Match clock segments: opening, mid-match, and last 20 seconds.

First build

  • Define zones and behavior labels in the scouting app.
  • Build simple rules: time in Loader zone, Toggle approaches, and blocks.
  • Surface a single label with confidence, then log whether it was useful.

Risk to manage

Robot detection is messy in standard V5RC. Keep the model based on visible behavior and scout notes, then add vision only where it proves reliable.

Why it wins: if the dashboard recognizes a defender or Toggle camper early, the team can switch from its default plan to a counter-plan before the match is half over.

Stretch Override Strategy Sim + RL Trainer Train policies in a 2D sim, then extract human strategy lessons for the team. Medium value / high risk

What Ethan builds

A simplified 2D Override simulator with objects, Toggles, Goals, scoring, robot routes, match clock, and basic opponent policies. The point is strategy discovery, not direct robot control.

Inputs

  • Verified scoring rules from the game manual and Q&A.
  • Robot timing data from practice: travel, intake, scoring, and turning.
  • Opponent archetypes from the scouting and classifier projects.

First build

  • Implement a deterministic sim with scripted robot policies.
  • Run thousands of simple games and summarize winning sequences.
  • Add PPO or Q-learning only after the hand-coded sim behaves correctly.

Definition of done

The sim produces strategy notes the team can test in practice, such as which Toggle to prioritize when yellow Pins cluster in one quadrant.

Why it wins: the award story is strong, and the strategic discoveries may be useful, but this should wait until the team has the scoring rules and timing data nailed down.

Roadmap

A practical season plan for Ethan.

Now to July 2026

Foundation

Set up PROS, LemLib, a clean Git repo, and a RobotEvents API sandbox. Build the first scouting data model before real Override events begin.

August to September 2026

Scouting v1

Launch event lookup, team pages, manual scouting forms, and basic Elo. Use shell ratings while the season has limited match data.

October to November 2026

Prediction layer

Add TrueSkill, alliance-pick boards, match odds, auto win-point rate, Skills history, and endgame reliability estimates.

December 2026 to January 2027

Driver assist prototype

Ship a V5 Brain screen mode that displays next-action suggestions from clock, Toggle state, target priority, and odometry. Keep the first version rule-based.

February to March 2027

Regional polish

Tune ratings from real events, add opponent tendencies, and practice with the assistant until the driver trusts it under pressure.

April to May 2027

Portfolio and awards

Package the build log, scouting results, dashboard screenshots, code architecture, and match examples into a judge-ready technical story.

Working Plan

The first two weeks should produce something the team can touch.

Week 1

  1. Create a RobotEvents API client and cache layer.
  2. Define team, event, match, skills, and scouting schemas.
  3. Build event search and registered-team pages.
  4. Add manual scout notes for auton, scoring style, defense, and endgame.

Week 2

  1. Add simple ratings and predicted match win percentage.
  2. Create an alliance-pick board for one real tournament.
  3. Export a print-friendly pick list for alliance selection.
  4. Start a small PROS field-state module for the future driver assistant.

Sources

Canonical links to keep pinned.

VEX current game Reveal thread Official Q&A Game manual Field assembly Field CAD LemLib SIGBots path planning RobotEvents API AI Vision Sensor

Updated May 23, 2026. Always verify exact rules, robot constraints, and Q&A rulings against the official manual before building competition-critical code.