NVIDIA Isaac Lab Guide 2026 — GPU-Accelerated Robot Learning

# Required:
# - Ubuntu 20.04 or 22.04
# - NVIDIA GPU: RTX 3070 minimum (RTX 3090 / A100 recommended)
# - CUDA 11.8 or 12.x
# - Python 3.10

# Verify GPU and CUDA
nvidia-smi
nvcc --version

# Method 1: NVIDIA Omniverse Launcher (GUI, recommended)
# Download from: developer.nvidia.com/isaac-sim
# Install Isaac Sim 4.x via the Launcher

# Method 2: pip (Isaac Sim 4.0+)
pip install isaacsim==4.2.0 \
    isaacsim-extscache-physics==4.2.0 \
    isaacsim-extscache-kit==4.2.0 \
    isaacsim-extscache-kit-sdk==4.2.0 \
    --extra-index-url https://pypi.nvidia.com

# Clone the Isaac Lab repository
git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab

# Install Isaac Lab in development mode
./isaaclab.sh --install

# Verify the installation
./isaaclab.sh -p source/standalone/tutorials/00_sim/create_empty.py

# Train ANYmal-C locomotion with PPO (classic first experiment)
./isaaclab.sh -p source/standalone/workflows/rsl_rl/train.py \
    --task Isaac-Velocity-Rough-Anymal-C-v0 \
    --num_envs 4096 \
    --headless

# Watch training progress
# Checkpoint saved to: logs/rsl_rl/anymal_c_rough/

# Play back the trained policy (with rendering)
./isaaclab.sh -p source/standalone/workflows/rsl_rl/play.py \
    --task Isaac-Velocity-Rough-Anymal-C-v0 \
    --num_envs 32 \
    --checkpoint logs/rsl_rl/anymal_c_rough/<run>/model_*.pt

from isaaclab.envs import ManagerBasedRLEnv

# Load a pre-built environment
env = gym.make("Isaac-Velocity-Rough-Anymal-C-v0",
               num_envs=4096,
               device="cuda:0")

# Step the environment
obs, info = env.reset()
for _ in range(1000):
    action = env.action_space.sample()  # Random policy
    obs, reward, terminated, truncated, info = env.step(action)

# Example: Custom reward manager config
from isaaclab.managers import RewardTermCfg as RewTerm
from isaaclab.managers import SceneEntityCfg

@configclass
class MyRewardsCfg:
    # Reward forward velocity
    track_lin_vel_xy_exp = RewTerm(
        func=mdp.track_lin_vel_xy_exp,
        weight=1.0,
        params={"command_name": "base_velocity", "std": 0.5}
    )
    # Penalty for joint torque
    dof_torques_l2 = RewTerm(
        func=mdp.joint_torques_l2,
        weight=-1.0e-5
    )

from isaaclab_assets import ANYMAL_C_CFG, UNITREE_GO2_CFG

# Configure ANYmal C
anymal_cfg = ANYMAL_C_CFG.replace(
    prim_path="{ENV_REGEX_NS}/Robot",
    spawn=sim_utils.UsdFileCfg(
        usd_path=f"{ISAACLAB_NUCLEUS_DIR}/Robots/ANYbotics/ANYmal-C/anymal_c.usd",
        rigid_props=sim_utils.RigidBodyPropertiesCfg(
            disable_gravity=False,
            max_depenetration_velocity=1.0,
        ),
    ),
)

from isaaclab.terrains import TerrainImporterCfg

terrain_cfg = TerrainImporterCfg(
    prim_path="/World/ground",
    terrain_type="generator",
    terrain_generator=ROUGH_TERRAINS_CFG,
    max_init_terrain_level=5,
    collision_group=-1,
    physics_material=sim_utils.RigidBodyMaterialCfg(
        friction_combine_mode="multiply",
        restitution_combine_mode="multiply",
        static_friction=1.0,
        dynamic_friction=1.0,
    ),
)

# Domain randomization config
@configclass
class MyEventsCfg:
    # Randomize base mass each episode
    add_base_mass = EventTerm(
        func=mdp.randomize_rigid_body_mass,
        mode="startup",
        params={
            "asset_cfg": SceneEntityCfg("robot", body_names="base"),
            "mass_distribution_params": (-1.0, 3.0),
            "operation": "add",
        },
    )
    # Push robot with random force
    push_robot = EventTerm(
        func=mdp.push_by_setting_velocity,
        mode="interval",
        interval_range_s=(10.0, 15.0),
        params={"velocity_range": {"x": (-0.5, 0.5), "y": (-0.5, 0.5)}},
    )