This repository holds the code used for our recent work on estimating physical parameters from robot interactions. Our experiments show how a robot can have a sense of consistent physical parameters even when only sparse noisy data is available.

I usually build pybullet from sources. In this case, you need the following information in your $PYTHONPATH variable:

$ export PYTHONPATH=$PYTHONPATH:$BULLET/build_cmake/examples/pybullet:$BULLET/examples/pybullet/gym

Assuming $BULLET is the directory where you cloned bullet repository.

In this case, you should be able to run:

$ python rippe_parallel.py

Which relies on joblib to parallelize the simulation execution and you can change the parameters in parametersConfig.py.

Warning: The file rippe_nonparallel.py was used for debugging and ignores the information in parametersConfig.py. You have to manually change the source for your desired parameters.

If you use any part of this program, we kindly ask you to cite the following related publication:

@inproceedings{dehban2019robotic,
  title={Robotic Interactive Physics Parameters Estimator (RIPPE)},
  author={Dehban, Atabak and Cardoso, Carlos and  Vicente, Pedro and Bernardino, Alexandre and Santos-Victor, Jos{\'e}},
  booktitle="IEEE International Conference on Developmental and Learning and on Epigenetic Robotics (ICDL-Epirob)",
  year={2019},
  note={in press}
}