Note
Code & Documentation coming soon.
We use extended_legged_gym as the major simulation environment.
1.Install virtual environment
mamba create -n pegasusflow python=3.8
mamba activate pegasusflow2.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.shNote
Still Work In Progress.
Run the demo with default settings:
python demos.py --robot anymal_cThe following robot tasks are available:
anymal_c: ANYmal-C quadruped robotelspider_air_barrier_nav: ElSpider robot with air barrier navigationelspider_air_timberpile_nav: ElSpider robot with timber pile navigationfranka: Franka robot arm
--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
ANYmal-C locomotion:
python demos.py --robot anymal_c --num_envs 1 --rollout_envs 128 --command walk_forwardElSpider barrier navigation:
python demos.py --robot elspider_air_barrier_nav --num_envs 2 --rollout_envs 128Franka arm manipulation:
python demos.py --robot franka --num_envs 1 --rollout_envs 128 --command reach_backwardLocomotion commands (for anymal_c, elspider variants):
walk_forward,walk_backwardstrafe_left,strafe_rightturn_left,turn_rightstop
Manipulation commands (for franka):
reach_forward,reach_backwardreach_left,reach_rightreach_up,reach_downhome
Run headless (no GUI):
python demos.py --robot anymal_c --headlessDisable trajectory optimization (policy only):
python demos.py --robot anymal_c --disable_trajectory_opt