OpenAI-Gym porting of the Multi-Goal environment used in the paper "Reinforcement Learning with Deep Energy-Based Policies" (Haarnoja et al., 2017) [arXiv] [GitHub]. The Multi-Goal environment is a bi-dimensional world, with continuous states, actions, and rewards. The agent (a particle) must reach one of the rewards. Typical algorithms fall into a local minimum and are not able to reach all the rewards.
Acknowledgments: if you use this package, please acknowledge the original work of Haarnoja et al. (2017).
Differences with original version: the original code is not compatible with OpenAI-Gym. Therefore, the main difference is the removal of the calls to obsolete packages, which have been replaced with calls to Gym classes. The package only requires Numpy and Matplotlib. Note that the method render() in this implementation is different from the original one. It is defined as render(mode='png', file_path='./multigoal.png') and it saves an image with the trajectory taken by the agent up to that point. Ideally it should be called at the end of the episode (see image). An additional parameter called step_limit has been defined at init time to avoid agents running indefinitely.
Example: to test the code you can try this example based on a random agent
import gym
env= gym.make('gym_multigoal:multigoal-v0', step_limit=500)
env.reset()
for _ in range(300):
observation, reward, done, _ = env.step(env.action_space.sample()) # take a random action
if(done): break
env.render(mode='png', file_path='./multigoal.png')
env.close()License
MIT License
Copyright (c) 2021 Massimiliano Patacchiola
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.