This repository contains the software and data for the ICRA2023 submitted article "NMPC for Deep Neural Network-Based Collision Avoidance exploiting Depth Images"
Thanks again for the files. I have 2 follow-up questions on the training dataset:
I tried to run depth_state_check_train.py, but ran into an issue where the training dataset, half_clutter_train.hdf5, only contains ['raw'], but not ['collision'].
It appears that the images in the training dataset have values in the range of thousands. For instance, for the first image in the ['raw'], the max value is 5000 and the min value is 2463. Are these values given in another unit apart from meters, e.g., millimeters? Since the depth_max is 5, so should these values be lesser than 5? It also appears that the image, img_gpu in __getitem__ of DataCollisionCheck is clamped between 0 and 1.
Side note: There is a need to change the parameters hdf5_train and hdf5_test in half_clutter.yaml to half_clutter_train.hdf5 and half_clutter_test.hdf5 respectively, so that it is compatible with the files in the link given in the README file.
wmh02240
commented
1 week ago
Hi,
Thanks again for the files. I have 2 follow-up questions on the training dataset:
depth_maxis 5, so should these values be lesser than 5? It also appears that the image,img_gpuin__getitem__ofDataCollisionCheckis clamped between 0 and 1.Side note: There is a need to change the parameters hdf5_train and hdf5_test in half_clutter.yaml to half_clutter_train.hdf5 and half_clutter_test.hdf5 respectively, so that it is compatible with the files in the link given in the README file.