/mabel

Manipulation Benchmarks for Learning

Primary LanguagePythonApache License 2.0Apache-2.0

MABEL

MAnipulation BEnchmarks for Learning

MABEL is a collection of environments/tasks for benchmarking manipulation, simulated with the Mujoco physics engine and wrapped in the OpenAI gym API. 'MABEL' constains following tasks

Pointing Pouring Relocation Zipping
Alt text Alt text Alt text Alt text

Getting Started

MABEL uses git submodules to resolve dependencies. Please follow steps exactly as below to install correctly.

  1. Ensure you have access these repositories - Franka_sim, Furniture_sim and Scene_sim

  2. Clone this repo with pre-populated submodule dependencies

$ git clone --recursive git@github.com:vikashplus/mabel.git
  1. Update submodules
$ cd mabel  
$ git submodule update --remote
  1. Add repo to pythonpath by updating ~/.bashrc or ~/.bash_profile
export PYTHONPATH="<path/to/mabel>:$PYTHONPATH"
  1. Follow install instructions from mjrl to get model free agents for MABEL
  2. To visualize an env using a random policy
MJPL python mabel/mabel_agents/mjrl/examine_policy.py -i mabel -e MableReachRandom-v0
  1. To visualize a trained mjrl agent's policy
MJPL python mabel/mabel_agents/mjrl/examine_policy.py -i mabel -e MableReachRandom-v0 -p <path to policy>