Simple code to demonstrate Multi-Agent Deep Reinforcement Learning by using Asynchronous & Impala Proximal Policy Optimization in Pytorch
The code follow algorithm in PPO's implementation on OpenAI's baseline and using newer version of PPO called Truly PPO, which has more sample efficiency and performance than OpenAI's PPO. Currently, I am focused on how to implement this project in more difficult environment (Atari games, MuJoCo, etc).
- Clean up the code
- Use Truly PPO
- Use IMPALA
- Add Tensorflow 2 version
- Add more complex environment
- Add more explanation
This project is using Pytorch and Tensorflow 2 for Deep Learning Framework, Gym for Reinforcement Learning Environment, and Ray for Asynchronous. Although it's not required, but i recommend run this project on a PC with GPU and 8 GB Ram
Make sure you have installed Ray and Gym.
You can use Pytorch
- Click here to install pytorch
Just clone this project into your work folder
git clone https://github.com/wisnunugroho21/asynchronous_impala_PPO.git
After you clone the project, run following script in cmd/terminal :
cd asynchronous_impala_PPO/discrete/pytorch
python3 ppo_sync.py
cd asynchronous_impala_PPO/discrete/pytorch
python3 ppo_impala.py
PPO is motivated by the same question as TRPO: how can we take the biggest possible improvement step on a policy using the data we currently have, without stepping so far that we accidentally cause performance collapse? Where TRPO tries to solve this problem with a complex second-order method, PPO is a family of first-order methods that use a few other tricks to keep new policies close to old. PPO methods are significantly simpler to implement, and empirically seem to perform at least as well as TRPO.
There are two primary variants of PPO: PPO-Penalty and PPO-Clip.
-
PPO-Penalty approximately solves a KL-constrained update like TRPO, but penalizes the KL-divergence in the objective function instead of making it a hard constraint, and automatically adjusts the penalty coefficient over the course of training so that it’s scaled appropriately.
-
PPO-Clip doesn’t have a KL-divergence term in the objective and doesn’t have a constraint at all. Instead relies on specialized clipping in the objective function to remove incentives for the new policy to get far from the old policy.
OpenAI use PPO-Clip
You can read full detail of PPO in here
Proximal policy optimization (PPO) is one of the most successful deep reinforcement-learning methods, achieving state-of-the-art performance across a wide range of challenging tasks. However, its optimization behavior is still far from being fully understood. In this paper, we show that PPO could neither strictly restrict the likelihood ratio as it attempts to do nor enforce a well-defined trust region constraint, which means that it may still suffer from the risk of performance instability. To address this issue, we present an enhanced PPO method, named Truly PPO. Two critical improvements are made in our method: 1) it adopts a new clipping function to support a rollback behavior to restrict the difference between the new policy and the old one; 2) the triggering condition for clipping is replaced with a trust region-based one, such that optimizing the resulted surrogate objective function provides guaranteed monotonic improvement of the ultimate policy performance. It seems, by adhering more truly to making the algorithm proximal - confining the policy within the trust region, the new algorithm improves the original PPO on both sample efficiency and performance.
You can read full detail of Truly PPO in here
In this work we aim to solve a large collection of tasks using a single reinforcement learning agent with a single set of parameters. A key challenge is to handle the increased amount of data and extended training time. We have developed a new distributed agent IMPALA (Importance Weighted Actor-Learner Architecture) that not only uses resources more efficiently in single-machine training but also scales to thousands of machines without sacrificing data efficiency or resource utilisation. We achieve stable learning at high throughput by combining decoupled acting and learning with a novel off-policy correction method called V-trace. We demonstrate the effectiveness of IMPALA for multi-task reinforcement learning on DMLab-30 (a set of 30 tasks from the DeepMind Lab environment (Beattie et al., 2016)) and Atari-57 (all available Atari games in Arcade Learning Environment (Bellemare et al., 2013a)). Our results show that IMPALA is able to achieve better performance than previous agents with less data, and crucially exhibits positive transfer between tasks as a result of its multi-task approach.
You can read full detail of Impala in here
| Result Gif | Award Progress Graph |
|---|---|
 |
 |
| Result Gif |
|---|
 |
| Result Gif | Award Progress Graph |
|---|---|
 |
 |
If you using heavy or complex Environment, there is a possibility that the training process will cause a memory leak. I still looking for any solution for this.
This project is far from finish and will be improved anytime . Any fix, contribute, or idea would be very appreciated