Datasets:

pranavsaroha
/

aic-cheatcode-ds-v1-smoke

Expert demonstrations of optical cable insertion using the UR5e robot in the AI for Industry Challenge simulation. Demonstrations were generated using the CheatCode policy (ground-truth TF transforms). v1 dataset adds force/torque sensor data, joint velocities/efforts, contact flag, structured task metadata, and uses the commanded pose target as the action signal.

Episodes: 15
Frames: 6260
FPS: 20
Cameras: left, center, right (256x288, AV1)
Action space: absolute Cartesian pose target = [pose.x, pose.y, pose.z, pose.qx, pose.qy, pose.qz, pose.qw]
State space (46 dims):
- TCP pose (7): position xyz + quaternion xyzw
- TCP velocity (6): linear xyz + angular xyz
- Joint positions (7)
- Joint velocities (7)
- Joint efforts (7)
- Wrist wrench (6): force xyz + torque xyz (post-tare)
- F/T tare offset (6): the bias removed by the tare service
Per-frame off-limit contact flag (1 dim): 1.0 if the gripper or arm contacted an off-limit zone (enclosure walls, task board frame) within ±25ms of the frame, else 0.0.
Episode metadata includes structured task fields: cable_type, plug_type, port_type, port_name, target_module_name, time_limit, and the original trial key.

Robot

Universal Robots UR5e arm
Robotiq Hand-E gripper
ATI AXIA80-M20 6-axis force/torque sensor at the wrist
3 Basler acA2440-20gc wrist cameras with Edmunds 58-000 lenses (1152x1024 native, downsampled to 256x288 for training)

Controller

The policy outputs 7-dim absolute pose targets. At submission time, these are wrapped in MotionUpdate messages with constant impedance:

target_stiffness diag: [90, 90, 90, 50, 50, 50]
target_damping diag: [50, 50, 50, 20, 20, 20]
wrench_feedback_gains_at_tip: [0.5] * 6
feedforward_wrench_at_tip: 0
trajectory_generation_mode: MODE_POSITION

These constants match what CheatCode used during data collection (verified by inspecting recorded MotionUpdate messages — all values constant across all 1590 messages of a representative trial).

Source

Source runs: 5
Generator script: ~/generate_aic_configs.py
Collection script: ~/collect_all.sh
Bag-to-LeRobot converter: ~/bag_to_lerobot.py
AIC toolkit: https://github.com/intrinsic-dev/aic

Downloads last month: 125