Sim Data Collection Pipeline

To collect simulation data efficiently, our workflow uses a standardized Python script to automatically execute motion sequences from a specified source in Isaac Sim. The script is highly configurable, supporting a wide range of parameters to seamlessly control different robots, motion datasets, and simulation settings for your experiments.

For example, the following script enables simulation runs for the H1-2 robot using AMASS motion files and custom configurations:

${ISAACSIM_PATH}/python.sh scripts/run_simulation.py \
    --robot-name h1_2 \
    --motion-source amass \
    --motion-files motion_files/h1_2/amass \
    --valid-joints-file configs/h1_2_valid_joints.txt \
    --output-folder output \
    --fix-root \
    --physics-freq 200 \
    --render-freq 200 \
    --control-freq 50 \
    --headless
This script makes it straightforward to batch-process motion collections across assets and scenarios, enabling comprehensive and repeatable data gathering for sim-to-real analysis and model training.