GitHub - ricsonc/ActiveVisionManipulation: Reinforcement Learning of Active Vision for Manipulating Objects under Occlusions (original) (raw)

Instructions on how to use this repository:

Setup:

1. install mujoco 150
2. install the requirements from the "requirements" file in scripts/
3. download the weights for the matterport RCNN at https://www.dropbox.com/s/wv49pvnplxu900v/mask_rcnn_mujoco_0009.h5?dl=0
4. move them into the directory HER/rcnn/logs/mujoco20181222T1549/

Viewing results:

1. view cam.mp4 and vis.mp4 of each directory inside the saved_weights directory, which shows the view of the camera from each trained policy

Running trained models:

1. go to the scripts/ directory
2. comment out the appropriate block of code in results.sh and run it