Challenges

Sponsored by American Tractor

Run the Fleet

Questions (1)

Updated 15 Nov, 12:20 MST

Coordinate a team of autonomous tractors, and adapt when things go wrong
Presented by American Tractor Company (ATC)

The problem
Autonomous tractors can do field work on their own, but someone still has to decide which tractor does which job, in what order, and what happens when a machine breaks down or a route is blocked. Today a lot of that coordination is done by hand. Better coordination means more work gets done, with less idle time and fewer mistakes.

Your task
Build a coordinator that assigns jobs to a small fleet of simulated autonomous tractors and re-plans when conditions change. It can be agent-based (for example, using a language model) or algorithmic.

What you get

Prefer to build your own simulation? That’s fine.
Use any language and keep it simple:

  • The farm: 2 to 3 tractors and 5 to 10 fields on a grid or map.
  • Tractors: each has its own capabilities, a compatible implement (such as a sprayer or mower) and a fuel or battery level.
  • Fields and jobs: each field has a location, so you can work out travel time. Each job has a type, a field, a priority and a deadline.
  • Events: a way to trigger a breakdown. For stretch goals, also a blocked route or a changed priority.

Judges score your coordinator, not your simulator, so don’t spend too long on it.

Suggested architecture
Focus on the “AI Farm Orchestrator” layer of the diagram:

  • Fleet Agent: tracks tractor status and assigns jobs. This is the core of the challenge.
  • Navigation Agent: plans routes between fields.
  • Safety Agent: checks a plan before it runs.
  • Agronomy Agent (optional): field and crop decisions.

Add a simple dashboard so the operator can see progress (the “Command Center” in the diagram). You can skip the hardware layers: edge controller, cloud platform and real tractors.

Must-haves (minimum to submit)

  1. Take a job list: an operator submits a set of jobs.
  2. Assign compatible tractors: respect each tractor’s capabilities, implements, fuel or battery, and the job deadlines.
  3. Show progress: use a dashboard or map view.
  4. Recover from a breakdown: when a tractor fails, reassign its unfinished work without creating conflicting assignments or losing track of completed work.
  5. Explain the new plan: give a short, plain-language explanation of what changed.

Stretch goals

  • Handle blocked routes and changed priorities, not just breakdowns.
  • Plan around fuel or battery levels.
  • Beat a simple baseline (assign jobs in order, first come, first served) with less travel and less idle time.
  • Let the operator ask “why?” about any assignment.
  • Connect your coordinator to AgOpenWeb.

What to hand in (by 5:00 PM)

  • A link to your code
  • A short README: how to run it, and whether you used AgOpenWeb or your own simulation
  • Your scenario or configuration file (tractors, fields and jobs)
  • The event log from your demo run
  • A comparison against the simple baseline: jobs completed, travel time and idle time

Winners

  1. 1stFarm ManAIgementTeam 11 · Team5
  2. 2ndFleetForemanTeam 24 · CHOPPA

How you will be judged

  • Constraint Compliance (4 pts)○ 4 - Outstanding: No violations in any scenario ○ 3 - Good: Rare, minor violations ○ 1 - Frequent4 of 20 points
  • Completed Work (4 pts)○ 4- Outstanding: More useful work than the starter dispatcher ○ 2 - Good: On par ○ 1 - Less4 of 20 points
  • Adaptation to Events (3 pts)○ 3 - Outstanding: Replans fast, no lost or duplicated work ○ 2 - Good: Recovers ○ 1 - Fails to recover3 of 20 points
  • Operational Efficiency (3 pts)○ 4 - Outstanding: Less travel and idle time than the starter ○ 2 - Good: On par ○ 1 - Wasteful4 of 20 points
  • Explanations & Demo (2 pts)○ 5- Outstanding: Clear plan explanations and demo ○ 3 - Good: Adequate ○ 1 - Unclear5 of 20 points

Tiebreaker: "Would a farm operator trust this plan?" (1-5): ____ Consider: clear explanations, edge cases handled, less manual coordination. It is used only when two teams are level, and is never added to the total.

Each judge scores every row from 1 to 5. A team’s result is the plain average of its judges’ totals.

The organizers make the final decision.