Quickstart

Quickstart#

This page takes you from a fresh checkout to training, playing back, and inspecting a task. Run every command from the Isaac Lab repository root with uv run; uv creates and manages the project environment for you.

Install and run your first task#

Install uv, clone Isaac Lab, and enter the checkout:

curl -LsSf https://astral.sh/uv/install.sh | sh
git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab

Train Cartpole with the Newton MJWarp backend and open the Newton visualizer:

uv run isaaclab train --rl_library rsl_rl \
   --task Isaac-Cartpole-Direct --num_envs 16 --max_iterations 10 \
   physics=newton_mjwarp --viz newton

Training outputs, including checkpoints, are written under logs/. Use --help after any command to see its arguments:

uv run isaaclab train --help

Hint

uv run installs core dependencies automatically. When a command needs an optional integration, add --extra <name> before isaaclab. Pass a comma-separated list to enable several extras. For example:

uv run --extra ovphysx isaaclab train --rl_library rsl_rl \
   --task Isaac-Cartpole-Direct physics=ovphysx

Extras install capabilities; task selectors choose how to use them. For example, --extra ovphysx makes the OV PhysX integration available, while physics=ovphysx selects it for the task. Extras can be combined freely, and --extra all installs every backend, RL library, and visualizer at once. See Optional extras for the complete list.

Choose an RL library#

Pass --rl_library to train and play to choose the learning framework. Start with rsl_rl: it is included in the default uv run environment and is a good choice for most GPU-based training.

--rl_library

Best starting point

Run it with

rsl_rl

Fast GPU training and policy distillation.

uv run isaaclab train --rl_library rsl_rl ...

rl_games

PPO, SAC, and A2C training.

uv run --extra rl-games isaaclab train --rl_library rl_games ...

skrl

A broad algorithm selection, with PyTorch and JAX support.

uv run --extra skrl isaaclab train --rl_library skrl ...

sb3

Stable-Baselines3 workflows, including CPU-oriented experiments.

uv run --extra sb3 isaaclab train --rl_library sb3 ...

rlinf

VLA model fine-tuning.

uv run --extra rlinf isaaclab train --rl_library rlinf ...

The libraries support different algorithms, workflows, and tasks. See Reinforcement Learning Library Comparison for the full comparison.

The four commands to know#

All task commands accept --task <task_name>. Start by listing the tasks available in your installation:

uv run python scripts/environments/list_envs.py

Command

Use it to

Example

train

Train a policy with an RL library.

uv run isaaclab train --rl_library rsl_rl --task Isaac-Cartpole-Direct

play

Run a trained policy from a checkpoint.

uv run isaaclab play --rl_library rsl_rl --task Isaac-Cartpole-Direct --checkpoint latest

zero_agent

Check a task using zero actions; useful for confirming that it launches.

uv run isaaclab zero_agent --task Isaac-Cartpole-Direct --viz newton

random_agent

Check a task using random actions; useful for a quick interaction smoke test.

uv run isaaclab random_agent --task Isaac-Cartpole-Direct --viz newton

The play command can also select a specific run with --load_run and a checkpoint with --checkpoint. See Reinforcement Learning Workflows for the complete training and playback reference.

Choose a backend#

Add physics=<backend> to a task command to select its physics backend. For camera tasks, you can also choose a renderer backend with renderer=<backend>. The backends available to a task depend on its configuration; use the task help to see the supported selectors:

uv run isaaclab train --task Isaac-Cartpole-Direct --help

Selector

Backend

Required extra

physics=newton_mjwarp

Newton using the MuJoCo-Warp solver. This is a good default for the quickstart.

None

physics=newton_kamino

Newton using the Kamino solver. This backend is beta and supports a limited set of tasks.

None

physics=ovphysx

OV PhysX.

ov or ovphysx

physics=isaacsim_physx

Isaac Sim PhysX.

isaacsim

renderer=newton_renderer

Newton Warp renderer.

None

renderer=ovrtx

OV RTX renderer.

ov or ovrtx

renderer=isaacsim_rtx

Isaac Sim RTX renderer.

isaacsim

renderer=rtx

Automatic RTX renderer selection.

isaacsim or ovrtx

Add task-specific options with presets=<name>; for example:

uv run isaaclab train --rl_library rsl_rl \
   --task Isaac-Cartpole-Camera-Direct \
   physics=newton_mjwarp renderer=newton_renderer presets=rgb

See Backends and Presets for backend and preset selection, and Hydra Configuration System for arbitrary configuration overrides.

Visualize a task#

Use --viz (or --visualizer) to select one or more visualizers during training or playback. Pass a comma-separated list without spaces, such as --viz newton,rerun.

Option

Use it to

Required extra

--viz newton

Open the Newton visualizer.

None

--viz rerun

Stream the task to the Rerun visualizer.

rerun

--viz viser

Open the web-based Viser visualizer.

viser

--viz kit

Open the Kit visualizer when it is available in your environment.

isaacsim

Omit --viz or use --viz none

Run headlessly.

None

For example, view the same task in both the Newton and Rerun visualizers:

uv run --extra rerun isaaclab random_agent --task Isaac-Cartpole-Direct \
   physics=newton_mjwarp --viz newton,rerun

See Visualization for visualizer setup and configuration.

Replay a checkpoint#

Train Cartpole to create a checkpoint:

uv run isaaclab train --rl_library rsl_rl \
   --task Isaac-Cartpole-Direct --num_envs 16 --max_iterations 10 \
   physics=newton_mjwarp

Then replay the newest checkpoint in the Newton visualizer:

uv run isaaclab play --rl_library rsl_rl \
   --task Isaac-Cartpole-Direct physics=newton_mjwarp \
   --checkpoint latest --viz newton

Choose a checkpoint with one of the following selectors:

Selector

Loads

--checkpoint <path>

The checkpoint at the specified local path.

--checkpoint best

The library-specific best or final checkpoint. If none was saved separately, this resolves to latest.

--checkpoint latest

The highest-step checkpoint from the newest compatible run.

Next steps#