CaDeLaC Datasets

Datasets used to train the models for the Context-Aware Deep Lagrangian Networks for Model Predictive Control (CaDeLaC).

Panda Datasets

• panda_mj_101_rand_envs_20_runs_50Hz_lqr: 100 environments with randomly sampled loads at the end-effector, along with one environment with only the robot. Each environment contains 20 runs with random initialization and joint references, for a total of 2020 runs.
• panda_mj_nominal_env_20_runs_50Hz_lqr: 20 runs of a single environment only with the robot without any load.

If you find this dataset useful, please consider citing:

@misc{schulze2025_cadelac,
      title={Context-Aware Deep Lagrangian Networks for Model Predictive Control}, 
      author={Lucas Schulze and Jan Peters and Oleg Arenz},
      year={2025},
      eprint={2506.15249},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2506.15249}, 
}

For more information, please refer to the code repository: https://github.com/Schulze18/cadelac.

License

MIT