Recommended way to set up scene with large number of entities

[ac-93]

Hi, thanks for the hard work with Genesis!

Is there a recommended method for simulating a scene with a relatively high number of entities added (100+).

I've experimented with some simple approaches but tend to hit unstable simulation and segmentation faults as I increase the number of entities added. Here is some example code (modified from the nonconvex_mesh.py example). Using sphere's for the example but ideally these would be concave mesh morphs.

import argparse
import numpy as np
import genesis as gs


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("-v", "--vis", action="store_true", default=False)
    parser.add_argument("-c", "--cpu", action="store_true", default=False)
    args = parser.parse_args()

    ########################## init ##########################
    gs.init(backend=gs.cpu if args.cpu else gs.gpu)

    ########################## create a scene ##########################
    scene = gs.Scene(
        viewer_options=gs.options.ViewerOptions(
            camera_pos=(3.5, 0.0, 2.5),
            camera_lookat=(0.0, 0.0, 0.5),
            camera_fov=40,
        ),
        show_viewer=args.vis,
        rigid_options=gs.options.RigidOptions(),
    )

    ########################## entities ##########################
    plane = scene.add_entity(
        gs.morphs.Plane(),
    )

    tank = scene.add_entity(
        gs.morphs.Mesh(
            file="meshes/tank.obj",
            convexify=False,
            scale=5.0,
            fixed=True,
            euler=(90, 0, 0),
        ),
    )

    # Grid dimensions and spacing
    n, m, k = 2, 2, 25
    spacing = 0.15
    init_pos = np.array((0.0, 0.0, 1.0))

    # Generate grid of positions centered around zero
    x = np.linspace(-(n - 1) * spacing / 2, (n - 1) * spacing / 2, n)
    y = np.linspace(-(m - 1) * spacing / 2, (m - 1) * spacing / 2, m)
    z = np.linspace(0, k * spacing, k)
    x, y, z = np.meshgrid(x, y, z, indexing="ij")

    # Compute positions
    positions = np.stack((x, y, z), axis=-1).reshape(-1, 3) + init_pos

    # Add entities to the scene
    parts = [
        scene.add_entity(
            morph=gs.morphs.Sphere(pos=pos, radius=0.06, fixed=False),
        )
        for pos in positions
    ]

    B = 1
    scene.build(n_envs=B, env_spacing=(1.0, 1.0))

    for i in range(1000):
        scene.step()


if __name__ == "__main__":
    main()

[zhouxian]

if there are many entities, try truning on contact_island=True, which will greatly improve simulation performance

[ac-93]

use_contact_island=True did improve simulation speed but seems to be unrelated to the crashes. Increasing max_collision_pairs to better match the number of entities does stop the crashes.