junhyeokahn/pypnc

Python Implementation of Planning and Control

PyPnC

PyPnC is a python library designed for generating trajectories for a robot system and stabilizing the system over the trajectories.

Installation

• Install anaconda
• Clone the repository:
  $ git clone https://github.com/junhyeokahn/PyPnC.git
• Create a virtual environment and install dependancies:
  $ conda env create -f pypnc.yml
• Activate the environment:
  $ conda activate pypnc

Running Examples

Three Link Manipulator Control with Operational Space Control

• Run the code:
  $ python simulator/pybullet/manipulator_main.py

Atlas Walking Control with DCM planning and IHWBC

• Run the code:
  $ python simulator/pybullet/atlas_dynamics_main.py
• Send walking commands through keystroke interface. For example, press 8 for forward walking, press 5 for in-place walking, press 4 for leftward walking, press 6 for rightward walking, press 2 for backward walking, press 7 for ccw turning, and press 9 for cw turning.
• Plot the results:
  $ python plot/atlas/plot_task.py --file=data/history.pkl

Atlas Locomotion Planning with TOWR+

• For TOWR+, install additional dependancy ifopt
• Train a Composite Rigid Body Inertia network and generate files for optimization:
  $ python simulator/pybullet/atlas_crbi_trainer.py and press 5 for training
• Run TOWR+:
  $ mkdir build && cd build && cmake .. && make -j6 && ./atlas_forward_walk
• Plot the optimized trajectory:
  $ python plot/plot_towr_plus_trajectory.py --file=data/atlas_forward_walk.yaml --crbi_model_path=data/tf_model/atlas_crbi
• Replay the optimized trajectory with the robot:
  $ python simulator/pybullet/atlas_kinematics_main.py --file=data/atlas_forward_walk.yaml

Citation

@article{10.3389/frobt.2021.712239,
	author = {Ahn, Junhyeok and Jorgensen, Steven Jens and Bang, Seung Hyeon and Sentis, Luis},
	journal = {Frontiers in Robotics and AI},
	pages = {257},
	title = {Versatile Locomotion Planning and Control for Humanoid Robots},
	volume = {8},
	year = {2021}}

pypnc's Issues

Atlas example gets stuck

Not sure what I'm missing here but the Atlas example gets stuck after a few iterations.

I only know this because enabling the control computation time print out shows that it stops. This is not the case for the manipulator example.

Initialize
================================================================================
[WalkingState] STAND
ctrl computation time:  0.01023721694946289
ctrl computation time:  0.010233163833618164
ctrl computation time:  0.009650945663452148
ctrl computation time:  0.010210990905761719
ctrl computation time:  0.012571573257446289
ctrl computation time:  0.014132022857666016
ctrl computation time:  0.009562015533447266
ctrl computation time:  0.008954524993896484
ctrl computation time:  0.008904695510864258
ctrl computation time:  0.011704683303833008
ctrl computation time:  0.011358022689819336
ctrl computation time:  0.008925676345825195
ctrl computation time:  0.008788585662841797
ctrl computation time:  0.01023411750793457
ctrl computation time:  0.009508371353149414
ctrl computation time:  0.00990438461303711
ctrl computation time:  0.009284496307373047
ctrl computation time:  0.010796070098876953
ctrl computation time:  0.009235858917236328
ctrl computation time:  0.01134800910949707
ctrl computation time:  0.010422468185424805

Running on 20.04 LTS. Perhaps that's the issue.

Add reference to paper.

I think it's good add a reference to the paper so that people know what to cite when they use PyPnC to benchmark their whole-body controllers.

Ahn, J., Jorgensen, S. J., Bang, S. H., & Sentis, L. (2021). Versatile Locomotion Planning and Control for Humanoid Robots. Frontiers in Robotics and AI, 257.