Through the window, no autopilot

crazyflie-window-manualcrazyflieBitcraze Crazyflie 2 quadrotorManipulationhard

Reference solution, 470 of 1,400 steps.

Instruction

You fly a Bitcraze Crazyflie 2 nano-quadrotor (MuJoCo Menagerie model; 27 g and 9 cm across each) in an indoor flight arena. World frame in metres: +x east, +y north, +z up, floor at z = 0. The arena is 8 m x 6 m (x from -4 to 4, y from -3 to 3) and 3 m high, closed by walls and a ceiling. A full-height partition wall at x = 1.5 separates the west hall from the east room; the only way through it is a 0.36 m x 0.36 m window centred at (1.5, 1.8, 1.0). The west hall has a home pad, a teal pad A, three square racing gates (orange frames, 0.5 m openings) and three swarm pads S0, S1, S2; the east room has a yellow pad B and a 0.9 m high shelf with a white landing platform (0.3 m square) on top. Pads are 0.3 m squares. Every position is listed under arena in robo observe. Touching anything other than the floor, a pad or the platform is a crash and fails the task.

Task

Take off from the home pad in the west hall, fly through the window in the partition wall into the east room, and land on pad B at (3.0, 1.8). In this task the skills are disabled: fly with robo act velocity setpoints only, closing the loop yourself with robo observe.

Success: the drone rests on pad B, with no crash at any time. Judged by the episode server from the simulated state after you call robo done and the drones have held still for 10 steps. A drone "rests on" a pad or the platform when it touches it, moves slower than 0.1 m/s and is tilted less than 20 degrees.

Controls. robo act VX VY VZ YAW_RATE sets the velocity setpoint for the next step(s), each in [-1, 1]: VX, VY, VZ along world x, y, z (x 1 m/s) and YAW_RATE (x 90 deg/s, counter-clockwise seen from above). One step is 50 ms; --repeat N holds a setpoint for N steps. An onboard cascaded controller (velocity -> attitude -> thrust and body moments) tracks the setpoint. An all-zero action holds the current position; resting on a surface with VZ <= 0 the motors stay off, so take off with a positive VZ. robo skill is disabled in this task.

Observation. robo observe reports drones (one entry: pos, vel, yaw_deg, roll_deg, pitch_deg, landed_on (the pad or surface it rests on, else null) and motors_on), arena, progress where the task counts something, and safety_events. Cameras: cf0/chase (default, behind the drone), cf0/fpv (forward), overview (the west hall) and east_room.

The step budget is 1400 steps (70 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                          # 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 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.
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 done exactly once when finished.

Run this task

bench eval run \
  -d benchflow/crazyflie@0.1 \
  --registry https://robouse.ai/hub/registry.json \
  --agent oracle \
  --include crazyflie-window-manual

Pinned to robohub commit bcd0d002b9df. The verifier and the reference solution are not published.

No trials yet.