Supplementary materials for "Learning Quadrupedal Locomotion over Challenging Terrain"

Project page: https://leggedrobotics.github.io/rl-blindloco

Author: Joonho Lee (jolee@ethz.ch)

This repo contains

• trained policy networks for ANYmal c010 and c100 models.
  • c010 model is not trained with Slippery Hills terrain.
  • c100 is the latest model (added during the revision)
• ANYmal environments implemented in Raisim simulator.
  • Height map terrain is included
  • In the visualizer, press 'c' key to sample new command
  • In the visualizer, press space key to re-initialize the robot

Dependencies

• raisim (https://github.com/raisimTech/raisimlib)
• raisimOgre (https://github.com/raisimTech/raisimOgre)
• tensorflow-cpp (https://github.com/leggedrobotics/tensorflow-cpp)

Third-party

• anymal_c_simple_description(https://github.com/ANYbotics/anymal_c_simple_description)

Notes

• Some paths are hard-coded in test_c010.cpp and test_c100.cpp. Be caureful about them.
• This repository is not maintained anymore. If you have a question, send an email to jolee@ethz.ch.
• We don't take questions regarding installation. If you install the dependencies successfully, you can easily run this.
• For the codes in rsc/robot/, ANYbotics' license is applied. MIT license otherwise.
• Install the dependencies in you local build directory and link as follows.

cmake -DCMAKE_INSTALL_PREFIX=${YOUR_LOCAL_BUILD_DIRECTORY} -DCMAKE_BUILD_TYPE=Release ..

• You can install tensorflow-cpp locally as follows.

cd tensorflow/tensorflow
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=${YOUR_LOCAL_BUILD_DIRECTORY} -DCMAKE_BUILD_TYPE=Release ..
make install -j