Skip to content

MasterYip/PegasusFlow

Repository files navigation

PegasusFlow: Parallel Rolling-Denoising Score Sampling for Robot Diffusion Planner Flow Matching

Paper arXiv Project Page

Banner

Note

Code & Documentation coming soon.

We use extended_legged_gym as the major simulation environment.

Installation

1.Install virtual environment

mamba create -n pegasusflow python=3.8
mamba activate pegasusflow

2.Install IsaacGym

cd isaacgym/python
pip install -e .

3.Clone and install this repo

git clone https://github.com/MasterYip/PegasusFlow --recursive
cd PegasusFlow
bash ./install.sh

Getting Started (WIP)

Note

Still Work In Progress.

Basic Usage

Run the demo with default settings:

python demos.py --robot anymal_c

Available Robots

The following robot tasks are available:

  • anymal_c: ANYmal-C quadruped robot
  • elspider_air_barrier_nav: ElSpider robot with air barrier navigation
  • elspider_air_timberpile_nav: ElSpider robot with timber pile navigation
  • franka: Franka robot arm

Key Arguments

  • --num_envs: Number of main simulation environments (default: 1)
  • --rollout_envs: Number of rollout environments per main environment for trajectory optimization (default: 128)
  • --horizon_nodes: Number of control nodes in the planning horizon
  • --horizon_samples: Number of trajectory samples for optimization
  • --command: Motion command for the robot

Example Commands

ANYmal-C locomotion:

python demos.py --robot anymal_c --num_envs 1 --rollout_envs 128 --command walk_forward

ElSpider barrier navigation:

python demos.py --robot elspider_air_barrier_nav --num_envs 2 --rollout_envs 128

Franka arm manipulation:

python demos.py --robot franka --num_envs 1 --rollout_envs 128 --command reach_backward

Available Commands

Locomotion commands (for anymal_c, elspider variants):

  • walk_forward, walk_backward
  • strafe_left, strafe_right
  • turn_left, turn_right
  • stop

Manipulation commands (for franka):

  • reach_forward, reach_backward
  • reach_left, reach_right
  • reach_up, reach_down
  • home

Additional Options

Run headless (no GUI):

python demos.py --robot anymal_c --headless

Disable trajectory optimization (policy only):

python demos.py --robot anymal_c --disable_trajectory_opt

Acknowledgements

About

Implementation of PegasusFlow: Parallel Rolling-Denoising Score Sampling for Robot Diffusion Planner Flow Matching

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published