MotionBench

Every task has its own board

The Board

Like a lap-time board on a car show: on every task, the agents that got the whole job done on every run are ranked by time, quickest first. Everyone else sits below the line, ranked by how much of the goal they reached.

Simulated runs, for now. These 570 runs come from our simulator, made to show how results read and to test the scorer in public. They are not real robots or real people. Real runs replace them as teams send them in.

MotionBench

V5 waypoints

Simulated

Waypoint navigation on a 12 ft field. Drive through six waypoints in order, within 5 cm and 5°, in 20 seconds.

Fastest to the whole goal, on every run

  1. 1LLM-written auton, round 10AI-written10.98swhole goal, 10 of 10 runs
  2. 2LLM-written auton, round 3AI-written11.71swhole goal, 10 of 10 runs
  3. 3Team auton A (odometry + GPS correction)Team-written code12.75swhole goal, 10 of 10 runs
  4. 4Driver 05 with auto-alignHuman + assist13.19swhole goal, 10 of 10 runs
  5. 5Driver 12Human14.79swhole goal, 10 of 10 runs
  6. 6Driver 06Human15.13swhole goal, 10 of 10 runs
  7. 7Driver 10Human15.70swhole goal, 10 of 10 runs

Didn’t finish every run

  1. 8Driver 11Human97%of the goal · worst run 83%
  2. 9Driver 02Human95%of the goal · worst run 83%
  3. 10Driver 09Human95%of the goal · worst run 83%
  4. 11Driver 05Human88%of the goal · worst run 83%
  5. 12Driver 04Human85%of the goal · worst run 67%
  6. 13Driver 07Human83%of the goal · worst run 67%
  7. 14Team auton B (odometry only)Team-written code82%of the goal · worst run 17%
  8. 15Driver 01Human70%of the goal · worst run 50%
  9. 16Driver 08Human62%of the goal · worst run 50%
  10. 17Driver 03Human55%of the goal · worst run 50%
  11. 18LLM-written auton, round 1AI-written35%of the goal · worst run 0%
  12. 19Team auton C (timed, open loop)Team-written code20%of the goal · worst run 0%

A robot wandering at random gets 0% of the goal here.

MotionBench

V5 pick-up

Simulated

Collect and deliver on a 12 ft field. Collect six objects and deliver them to the corner goal zone in 60 seconds.

Fastest to the whole goal, on every run

  1. 1Driver 12Human37.14swhole goal, 5 of 5 runs
  2. 2LLM-written auton, round 10AI-written37.31swhole goal, 5 of 5 runs
  3. 3Driver 10Human37.95swhole goal, 5 of 5 runs
  4. 4LLM-written auton, round 3AI-written38.36swhole goal, 5 of 5 runs
  5. 5Driver 05 with auto-alignHuman + assist40.94swhole goal, 5 of 5 runs
  6. 6Team auton A (odometry + GPS correction)Team-written code42.95swhole goal, 5 of 5 runs
  7. 7Driver 02Human46.18swhole goal, 5 of 5 runs

Didn’t finish every run

  1. 8Driver 04Human97%of the goal · worst run 83%
  2. 9Driver 06Human97%of the goal · worst run 83%
  3. 10Driver 09Human97%of the goal · worst run 83%
  4. 11Driver 01Human93%of the goal · worst run 83%
  5. 12Driver 05Human93%of the goal · worst run 83%
  6. 13Driver 11Human93%of the goal · worst run 67%
  7. 14Driver 07Human87%of the goal · worst run 67%
  8. 15Driver 03Human83%of the goal · worst run 67%
  9. 16Driver 08Human83%of the goal · worst run 67%
  10. 17Team auton B (odometry only)Team-written code73%of the goal · worst run 50%
  11. 18Team auton C (timed, open loop)Team-written code37%of the goal · worst run 33%
  12. 19LLM-written auton, round 1AI-written30%of the goal · worst run 0%

A robot wandering at random gets 0% of the goal here.

MotionBench

AIM waypoints

Simulated

Waypoint navigation for VEX AIM on a 6 x 4 ft table. Drive a VEX AIM robot through five waypoints, within 3 cm and 10°, in 20 seconds.

Fastest to the whole goal, on every run

  1. 1LLM-written Python, round 3 (uses AI Vision)AI-written14.51swhole goal, 10 of 10 runs
  2. 2Student 04 with AI Vision auto-aimHuman + assist16.16swhole goal, 10 of 10 runs
  3. 3Student 07 (One Stick controller)Human16.16swhole goal, 10 of 10 runs
  4. 4Class code B (Python, AI Vision on tags and barrels)Team-written code16.91swhole goal, 10 of 10 runs
  5. 5Student 08 (One Stick controller)Human17.30swhole goal, 10 of 10 runs

Didn’t finish every run

  1. 6Student 01 (One Stick controller)Human98%of the goal · worst run 80%
  2. 7Student 03 (One Stick controller)Human96%of the goal · worst run 80%
  3. 8Class code A (Blocks, gyro and encoders)Team-written code94%of the goal · worst run 80%
  4. 9Student 09 (One Stick controller)Human82%of the goal · worst run 80%
  5. 10Student 02 (One Stick controller)Human72%of the goal · worst run 60%
  6. 11Student 05 (One Stick controller)Human66%of the goal · worst run 60%
  7. 12Student 04 (One Stick controller)Human60%of the goal · worst run 40%
  8. 13Student 06 (One Stick controller)Human58%of the goal · worst run 40%
  9. 14Student 10 (One Stick controller)Human56%of the goal · worst run 40%
  10. 15LLM-written Python, round 1AI-written34%of the goal · worst run 20%

A robot wandering at random gets 0% of the goal here.

MotionBench

AIM sorting

Simulated

Sort barrels by colour with VEX AIM. Sort three orange and three blue barrels into their colour zones in 45 seconds.

Fastest to the whole goal, on every run

  1. 1LLM-written Python, round 3 (uses AI Vision)AI-written36.57swhole goal, 5 of 5 runs
  2. 2Student 07 (One Stick controller)Human38.56swhole goal, 5 of 5 runs
  3. 3Student 04 with AI Vision auto-aimHuman + assist40.22swhole goal, 5 of 5 runs
  4. 4Student 08 (One Stick controller)Human41.20swhole goal, 5 of 5 runs
  5. 5Class code B (Python, AI Vision on tags and barrels)Team-written code42.23swhole goal, 5 of 5 runs

Didn’t finish every run

  1. 6Student 01 (One Stick controller)Human93%of the goal · worst run 83%
  2. 7Student 03 (One Stick controller)Human93%of the goal · worst run 83%
  3. 8Student 09 (One Stick controller)Human80%of the goal · worst run 67%
  4. 9Student 02 (One Stick controller)Human77%of the goal · worst run 67%
  5. 10Student 05 (One Stick controller)Human70%of the goal · worst run 67%
  6. 11Student 04 (One Stick controller)Human57%of the goal · worst run 33%
  7. 12Student 06 (One Stick controller)Human53%of the goal · worst run 33%
  8. 13Student 10 (One Stick controller)Human53%of the goal · worst run 33%
  9. 14Class code A (Blocks, gyro and encoders)Team-written code43%of the goal · worst run 33%
  10. 15LLM-written Python, round 1AI-written23%of the goal · worst run 17%

A robot wandering at random gets 0% of the goal here.

MotionBench

Room cleaning

Simulated

Room floor coverage. Sweep the free floor of a 4 × 3 m room with a 30 cm tool in 180 seconds.

Most of the goal, on average

    1. 1Robot vacuum, systematic (lane-planning)Commercial robot95%of the goal · worst run 95%
    2. 2Person 03 with stick vacuumHuman91%of the goal · worst run 90%
    3. 3LLM-written coverage planner, round 3AI-written90%of the goal · worst run 90%
    4. 4Person 02 with stick vacuumHuman88%of the goal · worst run 87%
    5. 5Person 01 with stick vacuumHuman87%of the goal · worst run 86%
    6. 6Person 05 with stick vacuumHuman87%of the goal · worst run 84%
    7. 7Person 06 with stick vacuumHuman83%of the goal · worst run 81%
    8. 8Person 04 with stick vacuumHuman81%of the goal · worst run 80%
    9. 9Robot vacuum, bump-and-turnCommercial robot59%of the goal · worst run 58%

    A robot wandering at random gets 49% of the goal here.

    Why times only for full finishes

    A fast run that leaves half the job undone isn’t a fast run. So a time only counts when the whole goal was reached, and only if that happened on every run, because the worst run is the one that loses a match. Times are the average over those runs.

    Robot vacuums have no finish line worth timing, so the room-cleaning board is ranked by how much of the floor goal was covered instead.

    Want the detail?

    The results page has the ranges, the worst and best run of every agent, and the full scoring. Every run can be replayed in the explorer.

    Put your robot on the board.