Model and Link to Other Repository

• for Active Continual Learning: https://github.com/MiaoDragon/Hybrid-MPNet
• as Active Continual Learning achieves better results than CMPNet, we release models for S2D and R3D environment here: https://drive.google.com/file/d/1p9l4-GayV4_YASXDSgPtqRtDNMIeff5p/view?usp=sharing

dataset

• For C2D dataset:
• For other dataset: please refer to the original MPNet repository at https://github.com/ahq1993/MPNet

Code Explaination:

• cmpnet_train.py, cmpnet_test.py: general training and testing pipeline
• data_loader_*.py: data loader for different env settings
• gem_eval.py: called by cmpnet_test.py to evaluate the model on the testing dataset
• plan_general.py: common plannning functions used in different envs
• plan_*.py: implements env-specific planning functions: Collision-Checker
• utility_*.py: implements normalization and unnormalization of data depending on envs
• plot_util.py: functions for plotting
• cmpnet_compare_sample.py: general pipeline to compare different sampling methods in GEM
• mpnet_train.py: End-2-End training MPNet without Continual Learning functionality
• Model: implements the neural network model for planner network and encoder network
  • GEM_end2end_model.py: implements the MPNet combined with Continual Learning functions (GEM)
  • model_*.py: different planner models
  • gem_utility.py: utility functions used to implement GEM
  • GEM_end2end_model_*.py: network implementing different sampling methods
  • AE: encoder network architecture
• exp: bash codes for training and testing
  • cmpnet_train*.sh: code for training for each envs in Continual Learning mode
  • cmpnet_test*.sh: code for testing for each envs in Continual Learning mode
  • compare_sampling_method*.sh: code to compare different sampling methods
  • mpnet_train*.sh: code for training without Continual Learning mode
  • mpnet_test*.sh: code for testing without Continual Learning mode

Instructions for reproduction:

• obtain dataset (Coming soon)
• change the data_path in the training/testing script under exp folder to the dataset path
• change the model_path in the training/testing script under exp folder to your desired path for saving the model, and important statistics
• run the bash file in exp folder to conduct experiments (GPU needed)

Instructions for C++ MPNet:

• make a new directory named deps, and download libtorch into it by following this obtained from pytorch website: e.g.: the following works for cuda 10.1 Download here (Pre-cxx11 ABI): https://download.pytorch.org/libtorch/cu101/libtorch-shared-with-deps-1.3.1.zip Download here (cxx11 ABI): https://download.pytorch.org/libtorch/cu101/libtorch-cxx11-abi-shared-with-deps-1.3.1.zip
• make a new directory called build in the main directory.
• in the build directory, run the following: cmake c++ make
• convert Python trained MPNet model to C++ by running py_model_to_cpp.py inside c++ folder.