/antfields-demo

Primary LanguageC

About

This is a minimal example of active learning ntfields with igibson setup.

Setup

  1. git clone this repo
  2. run docker build -t antfields:demo . under the root directory of this repo, once you built the docker image, you don't need to build it again unless you change the dockerfile.
  3. run docker run --rm -it -e DISPLAY=$DISPLAY --env="QT_X11_NO_MITSHM=1" -v /tmp/.X11-unix:/tmp/.X11-unix --volume="/home/exx/Documents/antfields-demo:/antfields" --gpus all antfields:demo /bin/bash to start the docker container. Note that you need to change the path to your own path. BVH library is only needed during evaluation.
  4. run python main.py to start the training. In main.py, you can switch mode. READ_FROM_COOKED_DATA: train model with previous collected data, EXPLORATION: train model with active learning(exploration) in gibson env.
# train model with default EXPLORATION mode
python main.py 

# train model with READ_FROM_COOKED_DATA mode
python main.py --no_explore

# evaluate model
python eval.py

Note

We use bvh-distance-queries library repo to calculate the distance between two objects. If the bvh library fails to install in dockerfile, you can install it manually by "cd /antfields/bvh-distance-queries && pip install -e .". Occupancy map is modified from repo to fit the igibson env. Data sampling and preprocessing is modified from repo