Isaac Gym Environments for Legged Robots
This repository provides the environment used to train ANYmal (and other robots) to walk on rough terrain using NVIDIA's Isaac Gym. It includes all components needed for sim-to-real transfer: actuator network, friction & mass randomization, noisy observations and random pushes during training.
Maintainer: Nikita Rudin
Affiliation: Robotic Systems Lab, ETH Zurich
Contact: [email protected]
Useful Links
Project website: https://leggedrobotics.github.io/legged_gym/ Paper: https://arxiv.org/abs/2109.11978
Installation
- Create a new python virtual env with python 3.6, 3.7 or 3.8 (3.8 recommended)
- Install pytorch 1.10 with cuda-11.3:
pip3 install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio==0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
- Install Isaac Gym
- Download and install Isaac Gym Preview 3 (Preview 2 will not work!) from https://developer.nvidia.com/isaac-gym
cd isaacgym_lib/python && pip install -e .- Try running an example
python examples/1080_balls_of_solitude.py - For troubleshooting check docs
isaacgym/docs/index.html)
- Install rsl_rl (PPO implementation)
- Clone https://github.com/leggedrobotics/rsl_rl
cd rsl_rl && git checkout develop && pip install -e .
- Install legged_gym
- Clone this repository
cd legged_gym && git checkout develop && pip install -e .
CODE STRUCTURE
- Each environment is defined by an env file (
legged_robot.py) and a config file (legged_robot_config.py). The config file contains two classes: one conatianing all the environment parameters (LeggedRobotCfg) and one for the training parameters (LeggedRobotCfgPPo). - Both env and config classes use inheritance.
- Each non-zero reward scale specified in
cfgwill add a function with a corresponding name to the list of elements which will be summed to get the total reward. - Tasks must be registered using
task_registry.register(name, EnvClass, EnvConfig, TrainConfig). This is done inenvs/__init__.py, but can also be done from outside of this repository.
Usage
- Train:
python issacgym_anymal/scripts/train.py --task=anymal_c_flat- To run on CPU add following arguments:
--sim_device=cpu,--rl_device=cpu(sim on CPU and rl on GPU is possible). - To run headless (no rendering) add
--headless. - Important: To improve performance, once the training starts press
vto stop the rendering. You can then enable it later to check the progress. - The trained policy is saved in
issacgym_anymal/logs/. Where/ _ /model_ .pt - The following command line arguments override the values set in the config files:
- --task TASK: Task name.
- --resume: Resume training from a checkpoint
- --experiment_name EXPERIMENT_NAME: Name of the experiment to run or load.
- --run_name RUN_NAME: Name of the run.
- --load_run LOAD_RUN: Name of the run to load when resume=True. If -1: will load the last run.
- --checkpoint CHECKPOINT: Saved model checkpoint number. If -1: will load the last checkpoint.
- --num_envs NUM_ENVS: Number of environments to create.
- --seed SEED: Random seed.
- --max_iterations MAX_ITERATIONS: Maximum number of training iterations.
- To run on CPU add following arguments:
- Play a trained policy:
python issacgym_anymal/scripts/play.py --task=anymal_c_flat- By default the loaded policy is the last model of the last run of the experiment folder.
- Other runs/model iteration can be selected by setting
load_runandcheckpointin the train config.
Adding a new environment
The base environment legged_robot implements a rough terrain locomotion task. The corresponding cfg does not specify a robot asset (URDF/ MJCF) and no reward scales.
- Add a new folder to
envs/with', which inherit from an existing environment cfgs_config.py - If adding a new robot:
- Add the corresponding assets to
resourses/. - In
cfgset the asset path, define body names, default_joint_positions and PD gains. Specify the desiredtrain_cfgand the name of the environment (python class). - In
train_cfgsetexperiment_nameandrun_name
- Add the corresponding assets to
- (If needed) implement your environment in
.py, inherit from an existing environment, overwrite the desired functions and/or add your reward functions. - Register your env in
isaacgym_anymal/envs/__init__.py. - Modify/Tune other parameters in your
cfg,cfg_trainas needed. To remove a reward set its scale to zero. Do not modify parameters of other envs!
Troubleshooting
- If you get the following error:
ImportError: libpython3.8m.so.1.0: cannot open shared object file: No such file or directory, do:sudo apt install libpython3.8
Known Issues
- The contact forces reported by
net_contact_force_tensorare unreliable when simulating on GPU with a triangle mesh terrain. A workaround is to use force sensors, but the force are propagated through the sensors of consecutive bodies resulting in an undesireable behaviour. However, for a legged robot it is possible to add sensors to the feet/end effector only and get the expected results. When using the force sensors make sure to exclude gravity from trhe reported forces withsensor_options.enable_forward_dynamics_forces. Example:
sensor_pose = gymapi.Transform()
for name in feet_names:
sensor_options = gymapi.ForceSensorProperties()
sensor_options.enable_forward_dynamics_forces = False # for example gravity
sensor_options.enable_constraint_solver_forces = True # for example contacts
sensor_options.use_world_frame = True # report forces in world frame (easier to get vertical components)
index = self.gym.find_asset_rigid_body_index(robot_asset, name)
self.gym.create_asset_force_sensor(robot_asset, index, sensor_pose, sensor_options)
(...)
sensor_tensor = self.gym.acquire_force_sensor_tensor(self.sim)
self.gym.refresh_force_sensor_tensor(self.sim)
force_sensor_readings = gymtorch.wrap_tensor(sensor_tensor)
self.sensor_forces = force_sensor_readings.view(self.num_envs, 4, 6)[..., :3]
(...)
self.gym.refresh_force_sensor_tensor(self.sim)
contact = self.sensor_forces[:, :, 2] > 1.
flying away.
5 Nov 5, 2022
2 Dec 6, 2021
277 Jan 5, 2023
70 Dec 7, 2022
259 Dec 11, 2022
406 Dec 26, 2022
184 Dec 31, 2022
99 Dec 12, 2022
74 Jan 3, 2023
185 Dec 26, 2022
1.2k Dec 25, 2022
32 Oct 10, 2022
25 Dec 10, 2021
17 Nov 3, 2022
20 Jul 18, 2022
2.8k Jan 3, 2023
535 Nov 15, 2022