Insert the held 8 mm peg into the socket's hole, all the way down

Insert the held 8 mm peg into the socket's hole, all the way down, when the measured hole position is off by a few millimetres

il-factory-peg-insert-noisygpu-isaaclabFranka PandaManipulationhard

Reference solution, 98 of 300 steps.

Instruction

The robot holds an 8 mm peg a few centimetres above a socket with an 8.1 mm hole (25 mm deep). Insert the peg into the hole and push it all the way down. The task is solved when the peg's base is within 2.5 mm of the hole's axis and within 1 mm of the bottom of the hole.

The measured position of the socket (fixed_part_position_estimate) is off by several millimetres in x and y (a fixed calibration error, the same all episode). Use the cameras and what happens when the peg touches the socket to find the hole.

The step budget is 300 steps: every robo act step and every robo skill call counts as one.

Scene and frames

World frame: metres, z up. The fixed part (hole, gear base or bolt) is bolted to the table; the robot already holds the other part a few centimetres above it, as Factory starts every episode. held_part_base is the pose of the held part's base (the peg's bottom end, the gear's hub, the nut's bottom face); fixed_part_position_estimate is where the fixed part's top centre is measured to be.

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; you cannot read or change the simulator, the scoring, or other files to succeed. The episode server checks Factory's own success condition on the final simulated state.

robo info                                   # action space, skills, cameras, step budget
robo observe                                # fingertip pose, held part pose, fixed part estimate and dimensions
robo observe --image [--camera C]           # also saves an RTX image (cameras: scene, close, side) and prints its path
robo act DX DY DZ RX RY RZ [--repeat N]     # one 1/15 s control step of Factory's task-space impedance controller
robo skill hold [N]                         # zero command for N steps (default 15)
robo done "short summary"                   # end the episode and ask for scoring
robo give-up "reason"                       # end the episode without claiming success

act is Factory's action: DX, DY, DZ in [-1, 1] set a position target up to 2 cm from the current fingertip position; RX, RY, RZ in [-1, 1] rotate the target by up to 0.097 rad. Roll and pitch are held level (the gripper keeps pointing down), so only RZ turns the hand. The target is clamped to within 5 cm of the fixed part, and actions are smoothed (an exponential moving average), so the hand responds over several steps. The controller is compliant: pushing against a surface builds force rather than moving the part.
Success is judged after you call robo done, with the robot holding still for 10 control steps. robo give-up, running out of steps or time, or stopping without robo done all score 0. Call robo done exactly once.

Run this task

robouse run \
  --task tasks/gpu-isaaclab/il-factory-peg-insert-noisy \
  --harness oracle \
  --out runs/il-factory-peg-insert-noisy

Pinned to robohub commit e472b1a1e041. The verifier and the reference solution are not published.

No trials yet.