Yueqian Liu

Hi @Mayankm96, thanks for the advice, I'll try to put it there!

So as I'm trying to add principal axes and moments of inertia attributes to the rigid body, I find that the function: https://github.com/NVIDIA-Omniverse/orbit/blob/fcc216a5399661b6f3125e1e0a99235f073a3124/source/extensions/omni.isaac.orbit/omni/isaac/orbit/sim/schemas/schemas.py#L302 lacks support for these settings. Any suggestion...

Hi @TheLonelyFighter, Thanks for commenting. I want to add that using the timestamp data that comes with `VehicleLocalPosition` (100Hz) is also fine (verified on real hardware). Cheers~

That works for me, many thx!

Thank you @Denys88 and @ViktorM for providing the examples and the pointer to the `A2CResnetBuilder`. While waiting for the answer, I was also looking at the code in [network builder](https://github.com/Denys88/rl_games/blob/master/rl_games/algos_torch/network_builder.py)...

> @ViktorM @Denys88 the example above assumes that you are using a frozen network. You can't optimise the weights of this network because `rl_games` has `torch.infernce()` context for doing running...

I am trying to do e2e learning to also optimize the CNN weights. Why does normalizing the input to a network with no grad break the compute graph? Could you...

郑博 @xuankuzcr 您好！我有类似的问题想请教一下 。 我使用了一个类似L515的传感器（噪声比较大，最大量程2.5m左右，标定结果在[这里](https://drive.google.com/file/d/1mE6I2LN7I9eVmZ7p1V2uY3DI5F9QDaRb/view?usp=drive_link)，且可以认为Lidar frame和cam frame是一致的），正在试图运行LIVO2在自己的[数据集](https://drive.google.com/file/d/1o9LirEnLpqiOlSnMlPIt9dwvlKuq4idY/view?usp=drive_link)上，但是不成功。 https://github.com/user-attachments/assets/3e2572a1-c23a-418f-8db5-9dec0cb16751 相机配置文件我是这么写的： ```yaml cam_model: Pinhole cam_width: 320 cam_height: 240 scale: 0.5 # 这是什么？没改 cam_fx: 227.46204253 cam_fy: 227.42116608 cam_cx: 159.92612471 cam_cy: 120.20147916 cam_d0: 0.03141088...

@xuankuzcr 感谢回复。想顺势追问几个问题： 1. 关于初始化：是否需要让系统完全静止，手持时的抖动是否会有不好的影响？ 2. 关于点云质量：如果在所有其他参数准确的情况下，这种点云能够work吗？我的理解是在点云匹配时平面很难找准，所以可能还是有点困难，对吗？如果可以work的话对于参数调节有什么其他建议么？ 3. 关于相机外参：点云是直接从深度图反投影得到的，这样TF应该是单位矩阵没错？然后图2是如何看出来是外参而不是点云本身质量问题导致的不和谐呢？ 4. 关于内参：d0到d3对应的是k1, k2, p1, p2吗？ 5. 关于其他提供点云的传感器：点云质量方面L515优于我这个数据集，那D435和D455之类的呢？这方面有进一步评论吗？ 谢谢您看到这里，问的问题或许有点naive，望指教～

> 1. 可以手抖，是初始化时间太短了，最好超过3s > 2. 平面厚度快半米了，不能work > 3. 原始的深度图和灰度图就没有对得很准，因为你反投影操恢复的有色点云棋盘格边缘就很不锐利 > 4. 是 > 5. 深度相机没有测过 谢谢解答～ @xuankuzcr