soldierofhell

Thanks @clydemcqueen. I think it's possible (depth + pose). I will write I manage to do this.

Thanks, this is exactly what I was asking, great survey.

Just a quick question: are you going to standarize dataset to detectron2 coco panoptic format? I'd like to add a2d2 to the current set. Regards,

Just noticed the same comment in your code :)

If I start decoding with zero I indeed receive "garbage" sometimes (usually a duplicate of first character), but if the same slicing is is in the cost function then it's...

In general, yes, it works. I find the documentation pretty comprehensive, but unfortunately seems like there's nobody here from Waymo to answer questions.

Also 'model' call should be preceded with `torch.no_grad()` context Edit: Ok, I found this is done before `eval_one_epoch` https://github.com/sshaoshuai/Pointnet2.PyTorch/blob/master/tools/train_and_eval.py#L176

Hi @croesmann, I'd like to start moving towards dynamic model with extending states with linear velocity and control with acceleration, like in kinematic model in [Kong](https://borrelli.me.berkeley.edu/pdfpub/IV_KinematicMPC_jason.pdf).  I'm not sure...

Also I noticed that currently xref in step() is only from goal. Shoudn't we include intermediate points of path in cost too? (smoothed?) https://github.com/rst-tu-dortmund/mpc_local_planner/blob/8a5980a7aedfc6385a2329a3f2baba8318091857/mpc_local_planner/src/controller.cpp#L129 https://github.com/rst-tu-dortmund/mpc_local_planner/blob/8a5980a7aedfc6385a2329a3f2baba8318091857/mpc_local_planner/src/controller.cpp#L169 I'd expect cost term similar...

Hi @amakarow, Thanks for the answer. I feel like everything is clear about adding extended model (I'll make a PR when finish). As for the global plan I realized that...