Waypoint following enviroment development

[PaoloGinefra]

Hi everyone, right now I am trying to develop an Isaac lab direct RL environment for a waypoint following task of a mobile robot. The task is simply to follow the waypoints without colliding with some obstacles that need to be randomized at every reset.

I am a bit stuck on how to spawn the obstacles and then on how to randomize their positions. My best try so far uses the Rigid Body Collection (even tho the obstacles need to be static) but I can't figure out how to spawn multiple obstacles per env.

Also rn I have a basic waypoint following implementation.

Any suggestions or resources are more than appreciated since I am new to this framework.

[RandomOakForest]

Thank you for posting this. To develop a waypoint following RL environment in Isaac Lab with randomized static obstacles, you can try to implement the following approaches:

Obstacle Spawning Methods

1. Rigid Object Collections

from isaaclab.assets import RigidObjectCfg
from isaaclab.sim import sim_utils

obstacle_cfg = RigidObjectCfg(
    prim_path="/World/envs/env_.*/Obstacles",
    spawn=sim_utils.MultiAssetSpawnerCfg(
        assets_cfg=[
            sim_utils.CuboidCfg(size=(0.5, 0.5, 1.0)),
            sim_utils.ConeCfg(radius=0.3, height=1.0),
            sim_utils.CylinderCfg(radius=0.4, height=1.2)
        ],
        random_choice=True,
        rigid_props=sim_utils.RigidBodyPropertiesCfg(
            solver_position_iteration_count=4,
            kinematic=True  # For static obstacles
        )
    )
)

This creates multiple obstacle types that spawn randomly in each environment while maintaining static behavior through kinematic properties.

2. Position Randomization at Reset

def _reset_obstacles(self, env_ids):
    # Generate random positions/orientations
    num_obstacles = 5  # Per environment
    positions = torch.rand((len(env_ids), num_obstacles, 3)) * 5.0
    orientations = torch.rand((len(env_ids), num_obstacles, 4)) * 2 - 1
    
    # Apply to all environments
    self._obstacles.set_world_poses(
        positions.reshape(-1, 3),
        orientations.reshape(-1, 4),
        env_indices=env_ids
    )

Call this during the reset phase to randomize obstacle positions while maintaining static object properties.

Recommended Implementation Strategy

1. Spawn Configuration:
   • Use MultiAssetSpawnerCfg to create diverse obstacle shapes
   • Set kinematic=True for static behavior
2. Collision Management:

collision_props=sim_utils.CollisionPropertiesCfg(
    collision_enabled=True,
    contact_offset=0.02,
    rest_offset=0.002
)

Ensure proper collision detection with robot sensors (see this example tutorial).