ganatrask/trossen_sim

Trossen Gripper Pick-and-Place Simulation Dataset

Simulated pick-and-place episodes for imitation learning, generated in MuJoCo with a Trossen parallel-jaw gripper on a 3-DOF Cartesian gantry.

Dataset Summary

Episodes	500
Steps per episode	145
Record frequency	20 Hz
Simulator	MuJoCo (timestep: 1ms)
Task	Randomized cube pick-and-place
Success rate	~100% (placement error < 20mm)

Task Description

A parallel-jaw gripper mounted on a 3-DOF (x, y, z) Cartesian gantry picks up a 2cm cube (30g) from a randomized position on a table and places it at a randomized target location. Each episode follows a 13-point waypoint trajectory with smooth interpolation.

File Structure

episodes/
  episode_0000.h5 ... episode_0499.h5
videos/
  wrist_cam_0000.mp4 ... wrist_cam_0499.mp4

HDF5 Episode Format

Each episode_NNNN.h5 contains:

Key	Shape	Dtype	Description
actions/ee_pose	(145, 7)	float32	Target end-effector pose [x, y, z, qw, qx, qy, qz]
actions/gripper	(145, 1)	float32	Gripper command (0.0 = open, 1.0 = closed)
observations/ee_pose	(145, 7)	float32	Actual end-effector pose [x, y, z, qw, qx, qy, qz]
observations/gripper_width	(145, 1)	float32	Gripper opening in meters (0 - 0.088m)
observations/timestamp	(145,)	float64	Seconds since episode start
videos/wrist_cam	(145,)	int64	Frame indices into companion MP4
env_state	scalar	JSON	Episode metadata (pick/place positions, placement error)
success	scalar	bool	True if placement error < 20mm
episode_length	scalar	int64	Number of timesteps (145)

Conventions

Convention	Value
Action frame	World (absolute)
Observation frame	World (absolute)
Quaternion format	wxyz (constant, no rotation DOF)
Gripper action	0 = open, 1 = closed

Simulation Parameters

Parameter	Value
Gantry workspace	X: ±0.25m, Y: ±0.15m, Z: -0.4m to 0m
Gripper stroke	0 - 88mm
Cube size	2cm, 30g
Gantry Kp	5000
Gripper Kp	500

Usage

import h5py

with h5py.File("episodes/episode_0000.h5", "r") as ep:
    actions = ep["actions/ee_pose"][:]       # (145, 7)
    gripper = ep["actions/gripper"][:]       # (145, 1)
    obs_pose = ep["observations/ee_pose"][:] # (145, 7)
    success = ep["success"][()]              # bool

Citation

If you use this dataset, please cite this repository.