Crate detour (no skills)
go2-crate-detour-manualquadrupedUnitree Go2Navigationmedium
Instruction
You control a Unitree Go2 quadruped (MuJoCo Menagerie model, 15 kg, 0.7 m long, base about 0.29 m above the floor when standing) in a small indoor plant. World frame in metres: +x east, +y north, +z up, floor at z = 0; headings are in degrees, 0 = facing +x (east), 90 = facing +y (north). The plant is 12 m x 8 m (x from -6 to 6, y from -4 to 4), closed by 1 m walls. A partition wall at x = 1 separates the west hall from the east plant room; its only opening is a 1.1 m wide doorway (y from 1.5 to 2.6) at the east end of a corridor (walls at y = 1.3 and y = 2.9, from x = -2.2 to the partition; open at its west end). The west hall has the blue dock (0.9 m square at (-4.8, -2.8)), a row of crates (0.6 m cubes, 0.8 m high, at x = -3.2 from y = -3.95 to -0.95) and one more crate at (-1.9, -1.1), a green mat at (-1.2, -2.8), and a pipe with a valve and an inspection tag on the west wall. The east room has a pump (radius 0.3 m, 0.6 m high, at (4.2, -2.6)) with an asset tag, an instrument board with a pressure gauge 1.4 m up on the east wall, a yellow panel with a red emergency-stop button and a green reset button on the east wall, and a 0.2 m high mezzanine platform in the north-east corner (x 4.4-6, y 1.6-4) with a control cabinet, reached by a ramp (x 2.8-4.4, y 2.4-3.4) that rises eastwards. Every fixed position is listed under facility in robo observe. The robot stands on a free-floating base: only its feet touching the floor move it, so it can slip, trip over obstacles or fall; a fall (base tilted more than 50 degrees, or the body touching the floor, the ramp or the platform) is recorded in safety_events and fails the task.
Task
The Go2 stands on the dock at (-4.8, -2.8), facing east. Walk to the green mat at (-1.2, -2.8) and stop on it, using only robo act velocity commands (skills are disabled). The row of crates stands between the dock and the mat, and one more crate stands north-west of the mat; plan a route around them.
Success: the base centre is within 0.35 m (horizontally) of the mat centre (-1.2, -2.8), at the end the robot is standing still (not stepping, base slower than 0.1 m/s and turning slower than 11 deg/s), and it never fell. Judged by the episode server from the simulated state after you call robo done and the robot has held still (zero action) for 10 steps; no new inspections or checkpoints are recorded during those 10 steps.
Controls. robo act VX VY WZ DHEIGHT DPITCH sets the action for the next step(s). VX and VY (m/s, up to +-0.6 and +-0.4) and WZ (deg/s, up to +-69, counter-clockwise seen from above) are the body-frame velocity command (x forward, y left) for the onboard trot controller (gait clock, foot placement from the commanded and measured velocity, leg inverse kinematics, attitude feedback and joint servos); all three zero means stop stepping and stand. DHEIGHT (m, +-0.02 per step) and DPITCH (deg, +-5 per step) change the body-height and nose-up pitch setpoints (height from -45 % to +5 % of the standing height, pitch +-20 degrees; a non-zero velocity command raises a lowered body to at least 85 % of the standing height); zero keeps them. One step is 50 ms; --repeat N holds an action for N steps. robo skill is disabled in this task.
Observation. robo observe reports robot (pos of the base centre, yaw_deg, roll_deg, pitch_up_deg, vel_body, yaw_rate_deg_s, gait = standing / walking / sitting, body_pose_setpoint, standing_height_m, feet_in_contact, the head_camera position and forward direction, and fallen), facility (every fixed landmark), objects (this task's movable objects and floor markings, if any), progress (inspected_tags, and the task's counters) and safety_events. Cameras: chase (default; follows the robot from above and behind-left), head (forward-facing on the front of the base, 70 degree field of view, square image), overview, west_hall and east_room.
The step budget is 900 steps (45 s).
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 # the robot, its sensors, action groups, skills and step budget
robo observe # robot and scene state as numbers
robo observe --image [--camera C] # also saves a camera image and prints its path (open it to look)
robo act V1 V2 ... [--repeat N] # one low-level action (the action groups under Controls), applied N times (N <= 50)
robo skill NAME ARG ... # run a skill listed by `robo info`; it runs until it finishes and reports the result
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 (+z up); angles are in degrees unless a field says otherwise.
- The episode has a fixed step budget (see
robo info); every simulated control step counts, including the steps a skill runs. - Skills are ordinary controllers: they can fail, stop early or be blocked by the scene. Read what they report and re-observe.
- Success is judged about 10 steps after you call
robo done, with the robot holding still (each action group's hold value: zero for velocity and delta commands, full brake for a car), so the goal must still be true when the robot stops. - Call
robo doneexactly once when finished.
Run this task
bench eval run \
-d benchflow/quadruped@0.1 \
--registry https://robouse.ai/hub/registry.json \
--agent oracle \
--include go2-crate-detour-manualPinned to robohub commit bcd0d002b9df. The verifier and the reference solution are not published.