active_extrinsic

This repository contains the source code of the paper Active Extrinsic Contact Sensing: Application to General Peg-in-Hole Insertion.

Directories

• codes_real_experiments: contains the source codes to run the real insertion experiments on ABB 120 robot.
• codes_training_n_visualization: contains the source codes for the training the tactile module and the reinforcement learning policy. It also contains the codes that generates the 3D visualization with the collected data.
• gtsam-project-python: is the package that contains the custom factors required for the project. Is should be installed with GTSAM to be used.
• TD3_model: contains the trained TD3 reinforcement learning insertion policy.
• weights: contains the trained tactile module convolutional neural network.

PREREQUISITES

• Python 3.6+ is required.
• GTSAM is required.

  • To install the wrap package via GTSAM:

    • Set the CMake flag GTSAM_BUILD_PYTHON to ON to enable building the Pybind11 wrapper.

    • Set the CMake flag GTSAM_PYTHON_VERSION to 3.x (e.g. 3.7), otherwise the default interpreter will be used.

    • You can do this on the command line as follows:

      cmake -DGTSAM_BUILD_PYTHON=ON -DGTSAM_PYTHON_VERSION=3.7 ..

  • Alternatively, you can install the wrap package directly from the repo, but you will still need to install GTSAM.

• To install the custom GTSAM python package required for the project:
  • In the 'gtsam-project-python' directory, create the build directory and cd into it.
  • Run cmake ...
  • Run make, and the wrapped module will be installed to the python directory in the top-level.
  • To install the wrapped module, simply run make python-install.