Robo Use

Fly to a waypoint

drone-waypointDroneSkydio X2 quadrotorNavigationmedium

Reference solution, 183 of 600 steps.

Instruction

You fly a Skydio X2 quadrotor (1.3 kg, about 0.6 m across; MuJoCo Menagerie model) over a small town. The drone starts sitting on the launch pad. World frame in metres: +x east, +y north, +z up, ground at z = 0. The town has five square landing pads (1.2 m, marked H), an east-west road and a north-south road, six trees (crowns about 1.4 m in radius around 3.6 m height), a 9 m telecom mast with an equipment box and an inspection tag, a 7 m tall wall (a warehouse side) north of the launch pad, a two-storey house with an upstairs bedroom window on its east wall and an antenna on its roof, and a plaza where a crowd of nine people stands. All positions are listed under landmarks in robo observe. Hitting anything other than the ground or a pad counts as a crash and fails the task.

Task

Take off and fly to the point (6.0, 4.0, 3.0), then hover there and call robo done.

Success: the drone is within 0.3 m of (6.0, 4.0, 3.0) and moving slower than 0.3 m/s after the 10-step hold, with no crash. Judged by the episode server after you call robo done and the drone has hovered for 10 steps.

Controls. robo act VX VY VZ YAW_RATE sets velocity setpoints for the next step(s), each in [-1, 1]: VX, VY horizontal velocity along world x, y (x 1 m/s), VZ vertical velocity (x 1 m/s), YAW_RATE heading rate (x 60 deg/s, positive turns counter-clockwise seen from above). One step is 50 ms; use --repeat N to hold a setpoint for N steps. A cascaded flight controller (velocity -> attitude -> rotor thrusts) keeps the drone level and tracks the setpoints; zero setpoints hover in place. robo move-to X Y Z --tol 0.2 flies to a position (it steers the drone's predicted stopping point, so it does not overshoot; one call runs at most 150 steps, so repeat it for long legs). The drone's forward camera looks along its heading (yaw_deg, measured from +x toward +y), pitched 15 degrees down; turn with YAW_RATE to point it.

Observation. robo observe reports pos, vel, yaw_deg, roll_deg, pitch_deg, altitude, landed (resting on the ground or a pad), landmarks and safety_events. robo observe --image --camera fpv saves what the forward camera sees; --camera chase (default) shows the drone from behind and above, --camera overview the whole town.

The step budget is 600 steps (30 s of flight).

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.

Drone note: in this task robo act takes the drone setpoints described under Controls instead of DX DY DZ GRIP, and robo grip is not available.

Run this task

bench eval run \
  -d benchflow/robouse-core@0.1 \
  --registry https://robouse.ai/hub/registry.json \
  --agent oracle \
  --include drone-waypoint

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

Trial
GPT-6 Astra · CodexSolved214 / 60036 sTrial Compare
Kimi K3 · Claude CodeSolved309 / 60042 sTrial Compare
Kimi K3 · mini-swe-agentSolved376 / 6005 minTrial Compare
GLM-5.3 · mini-swe-agentSolved374 / 6003 minTrial Compare
GLM-5.3 · Claude CodeSolved332 / 60061 sTrial Compare
GLM-5.3 · CodexSolved141 / 60031 sTrial Compare