Robo Use

Roll the ball through the maze to the goal

gymrobotics-pointmaze-medium-s3medium seed 3 variantGymnasium-RoboticsPoint massNavigationmedium

Reference solution, 346 of 1,000 steps.

Instruction

A small ball moves on the floor of a walled medium maze, pushed by a force you choose (adapted from Gymnasium-Robotics, PointMaze_Medium-v3, start/goal seed 3).

Goal: Roll the ball through the maze to the goal.

Success: the ball centre comes within 0.45 m of goal_pos. The episode ends as solved the moment this happens (you do not need to stop on the goal or call robo done afterwards).

In robo observe: agent_pos/agent_vel are the ball's position (m) and velocity (m/s), goal_pos is the goal (shown in red in camera images; the ball is green), and maze_rows is the maze layout, one string per row from the top (+y) row down: # is a wall block, . is free floor. cell_size is the width of one cell in metres, cell_center gives the world coordinates of the centre of cell (row, col), and agent_cell/goal_cell are the [row, col] cells the ball and goal are in. Walls fill whole cells; the ball (radius 0.1 m) cannot pass through them. Camera images are top-down with +x to the right and +y up.

This task uses a 2-D action. robo act FX FY [--repeat N] applies a force (each in [-1, 1]) to the ball along world x and y for N steps of 10 ms each; the ball keeps its momentum, so brake by pushing the other way. robo move-to and robo grip are disabled here, and the DX DY DZ GRIP form in the general instructions below does not apply.

How the robot is controlled and scored

You are controlling a simulated robot. Read the task below, then solve it by running the robo command in your shell (start with robo info and robo observe). Keep going until the task is done, then call robo done once. Do not stop to ask questions; there is no human to answer.

How to control the robot

You are the robot's policy. You act only through the robo command in your shell. There is no other way to move the robot, and you cannot read or change the simulator, the scoring, or other files to succeed; the episode server judges the final physical state itself.

robo info                         # action space, available skills, step budget
robo observe                      # robot and object state as numbers
robo observe --image              # also saves a camera image and prints its path (open it to look)
robo act DX DY DZ GRIP [--repeat N]   # low-level action, applied N times (N <= 50)
robo move-to X Y Z [--grip G]     # skill: move the gripper toward a point (if enabled for this task)
robo grip G [--steps N]           # skill: hold position and set the gripper (+1 close, -1 open)
robo done "short summary"         # end the episode and ask for scoring
robo give-up "reason"             # end the episode without claiming success
  • Positions are in metres in the world frame (x, y on the table plane, z up).
  • The episode has a fixed step budget (see robo info); every simulated step counts, including skills.
  • Unless the task says otherwise, success is judged about 10 steps after you call robo done, with the robot holding still, so the goal must still be true when the robot stops.
  • Work in small steps and re-observe after each motion. Call robo done exactly once when finished.

Run this task

bench eval run \
  -d benchflow/robouse-core@0.1 \
  --registry https://robouse.ai/hub/registry.json \
  --agent oracle \
  --include gymrobotics-pointmaze-medium-s3

Pinned to robohub commit 9e672aa1e7f0. The verifier and the reference solution are not published.

Trial
GPT-6 Astra · CodexSolved224 / 1,00054 sTrial Compare
Kimi K3 · Claude CodeSolved428 / 1,00030 sTrial Compare
GLM-5.3 · mini-swe-agentSolved699 / 1,00058 sTrial Compare
GLM-5.3 · Claude CodeSolved494 / 1,00067 sTrial Compare
Opus 5.5 · Claude CodeSolved441 / 1,00032 sTrial Compare
GLM-5.3 · CodexSolved719 / 1,00078 sTrial Compare
Kimi K3 · mini-swe-agentNot solved1,000 / 1,00012 minTrial Compare