[Bug]: Luisa batch rendering with partnet mobility object

[NakuraMino]

Bug Description

luisa renderer is able to render a partnet mobility object when i use the default gs.renderers.RayTracer but errors when I use gs.renderers.BatchRenderer(use_rasterizer=False).

Steps to Reproduce

If possible, provide a script triggering the bug, e.g.

# code snippet triggering the bug
import genesis as gs
import numpy as np
from matplotlib import pyplot as plt

BATCH_RENDERING = False

if BATCH_RENDERING:
    # batch rendering fails
    gs.init(backend=gs.gpu)

    scene = gs.Scene(
        rigid_options=gs.options.RigidOptions(enable_collision=False, gravity=(0, 0, 0)),
        show_viewer=False,
        renderer=gs.renderers.BatchRenderer(
            use_rasterizer=False
        ),
    )

    scene.add_light(
        pos=(0.0, 0.0, 5.0),
        color=(1.0, 1.0, 1.0),
        intensity=1.0,
        dir=(0.0, 0.0, -1.0),
    )
    

    plane = scene.add_entity(
        morph=gs.morphs.Plane(pos=(0.0, 0.0, -0.5)),
    )

    franka = scene.add_entity(
        gs.morphs.MJCF(file="xml/franka_emika_panda/panda.xml"),
    )

    camera1 = scene.add_camera(
        res=(640, 480),
        pos=(0.5, -3.0, 2.5),
        lookat=(0.5, 0.0, 0.2),
        fov=60,
        GUI=False,
    )

    camera2 = scene.add_camera(
        res=(640, 480),
        pos=(0.5, 3.0, 2.5),
        lookat=(0.5, 0.0, 0.2),
        fov=60,
        GUI=False,
    )

    obj = scene.add_entity(
        material=gs.materials.Rigid(friction=5.0),
        morph=gs.morphs.URDF(
            file='/path/to/partnet-mobility/45524/mobility.urdf',
            scale=0.5,
            pos=(0.8,0,0.4),
            quat = (1,0,0,0),
            fixed=True
            )
    )

    scene.build()
    scene.step()
    rgb, depth, seg, normals = scene.render_all_cameras(rgb=True)
    plt.imshow(np.concatenate((rgb[0].cpu().numpy(), rgb[1].cpu().numpy()), axis=1))
    plt.show()


else:
    # non batch rendering:
    gs.init(backend=gs.gpu)

    scene = gs.Scene(
        rigid_options=gs.options.RigidOptions(enable_collision=False, gravity=(0, 0, 0)),
        viewer_options=gs.options.ViewerOptions(
            res=(1920, 1080),
            camera_pos=(8.5, 0.0, 4.5),
            camera_lookat=(3.0, 0.0, 0.5),
            camera_fov=50,
        ),
        renderer=gs.renderers.RayTracer(  # type: ignore
            env_surface=gs.surfaces.Emission(
                emissive_texture=gs.textures.ImageTexture(
                    image_path="textures/indoor_bright.png",
                ),
            ),
            env_radius=15.0,
            env_euler=(0, 0, 180),
            lights=[
                {"pos": (0.0, 0.0, 10.0), "radius": 3.0, "color": (15.0, 15.0, 15.0)},
            ],
        ),
    )

    plane = scene.add_entity(
        morph=gs.morphs.Plane(pos=(0.0, 0.0, -0.5)),
    )

    franka = scene.add_entity(
        gs.morphs.MJCF(file="xml/franka_emika_panda/panda.xml"),
    )

    camera1 = scene.add_camera(
        res=(640, 480),
        pos=(0.5, -3.0, 2.5),
        lookat=(0.5, 0.0, 0.2),
        fov=60,
        GUI=False,
    )

    camera2 = scene.add_camera(
        res=(640, 480),
        pos=(0.5, 3.0, 2.5),
        lookat=(0.5, 0.0, 0.2),
        fov=60,
        GUI=False,
    )

    obj = scene.add_entity(
        material=gs.materials.Rigid(friction=5.0),
        morph=gs.morphs.URDF(
            file='/path/to/partnet-mobility/45524/mobility.urdf',
            scale=0.5,
            pos=(0.8,0,0.4),
            quat = (1,0,0,0),
            fixed=True
            )

    )

    scene.build()
    scene.step()
    rgb1, depth, seg, normals = camera1.render(rgb=True)
    rgb2, depth, seg, normals = camera2.render(rgb=True)
    
    plt.imshow(np.concatenate((rgb1, rgb2), axis=1))
    plt.show()

45524.zip

Expected Behavior

batch rendering returns the same output as when I use the default RayTracer.

On another note, the color of the object also changes when I use the raytracer vs default rasterizer, which I don't understand.

Screenshots/Videos

Relevant log output

By toggling the BATCH_RENDERING flag, you should be able to see the error output. The error I get is the following: 

[Genesis] [03:38:51] [INFO] Compiling simulation kernels...
[Genesis] [03:38:57] [INFO] Building visualizer...
Using raytracer
All Vulkan layers supported
free(): invalid next size (fast)

Environment

• OS: Ubuntu 20.04
• GPU/CPU 2x RTX 4090s, AMD Ryzen Threadripper PRO 5975WX 32-Cores
• GPU-driver version 550.90.07
• CUDA / CUDA-toolkit version 12.1

Release version or Commit ID

version 0.3.6

Additional Context

No response