tmrl
acme
Our great sponsors
tmrl | acme | |
---|---|---|
11 | 11 | |
422 | 3,370 | |
10.0% | 1.3% | |
6.3 | 5.8 | |
3 days ago | 10 days ago | |
Python | Python | |
MIT License | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
tmrl
-
Problem with Truncated Quantile Critics (TQC) and n-step learning algorithm.
Hi all! I'm implementing a TQC with n-step learning in Trackmania (I forked original repo from here: https://github.com/trackmania-rl/tmrl, my modified version here: https://github.com/Pheoxis/AITrackmania/tree/main). It compiles, but I am pretty sure that I implemented n-step learning incorrectly, but as a beginner I don't know what I did wrong. Here's my code before implementing n-step algorithm: https://github.com/Pheoxis/AITrackmania/blob/main/tmrl/custom/custom_algorithms.py. If anyone checked what I did wrong, I would be very grateful. I will also attach some plots from my last training and outputs from printed lines (print.txt), maybe it will help :) If you need any additional information feel free to ask.
- Training an unbeatable AI in Trackmania [video]
- Can you beat trackmania AI?
-
Python RL Environments on Windows
I don't know if it fits your need : https://github.com/trackmania-rl/tmrl
-
New to reinforcement learning.
Hi, if you are gonna train a deep RL algorithm on a real robot and you are a beginner, I suggest you try out tmrl. This will allow you to try out a readily available algorithm (Soft Actor-Critic) in real-time on a real video-game (TrackMania), as real-world-like proxy for all the concerns you will encounter on real robot, and to rather easily develop your own robot-learning pipeline from there for your own robot. The repo has a huge tutorial exactly for this purpose.
-
AI Learns Mario Kart Wii (Rainbow DQN)
I see, and how did you handle the simulator and dynamics? Did you "step" the game or did you capture screenshots with constant time interval in real-time? I am asking because tmrl uses the second option in TrackMania, which makes the approach generalizable to all video games including MarioKart, but so far we met no such success training CNNs with the Soft Actor-Critic familly. The setting is a bit more difficult because it uses continuous input inckuding the gas and break (I suppose you always send gas to the maximum?) and we don't use punishments for collisions and this kind of tricks, but still, if it works that well in your setting I think it should work similarly in ours.
-
Have you used any good DRL library?
I am very disappointed these guys don't cite tmrl :D
-
osu!
If you just want to make a Gym environment, focus on rtgym. You will want to have a way of retrieving observations, say, raw images for instance. You can do this with pywin32 as done here. You will also want to grab a reward signal, which will probably be the most challenging thing to do because you will have to compute that from screenshots since you don't have access to the game internals (in fact you could because the game is opensource, but I assume you don't want to go down this path). If there is something like a score counter that's not moving somewhere, I suggest you capture that and read the numbers individually with the 1-NN algorithm (which is done in the un-supported "trackmania nation forever" versions of the tmrl encironment btw if you need help doing that). Then you will need to input controls. I see that osu is controller by the mouse and keyboard, there are several solutions for that. You can use pywin32 I think, pyautogui, keyboard, and probably others. Once you have these basic building blocks, it will be fairly straightforward to use the rtgym API in order to build you real time Gym environment.
-
Is replay buffer can remove "done"?
Sometimes it is more than okay, it may be necessary. For instance in tmrl we do exactly that, because we are in a partially observable environment where we cannot say whether the next state will be terminal or not, and where what we try to actually optimize is an infinite sum of discounted rewards.
-
[P] DeepForSpeed: A self driving car in Need For Speed Most Wanted with just a single ConvNet to play ( inspired by nvidia )
Cool project. Shameless self-advertising here but you can use vgamepad to control the game with a virtual gamepad instead of key presses, which enables analog policies. We do this in TrackMania :)
acme
-
Fast and hackable frameworks for RL research
I'm tired of having my 200m frames of Atari take 5 days to run with dopamine, so I'm looking for another framework to use. I haven't been able to find one that's fast and hackable, preferably distributed or with vectorized environments. Anybody have suggestions? seed-rl seems promising but is archived (and in TF2). sample-factory seems super fast but to the best of my knowledge doesn't work with replay buffers. I've been trying to get acme working but documentation is sparse and many of the features are broken.
-
How much of a MuJoCo simulation or real life robot can you train on a 3090?
I'm training a few algorithms from Deepmind's acme library on some MuJoCo models and I'm wondering how long this will take to train and what it's going to do to my electric bill. Is a 3090 or two enough to train something to keep its balance, or do a task, or do I need to wait for the 8090 to come out?
-
Recomendations of framework/library for MARL
Recently dm-acme also added support for multi-agent environments. Acme: https://github.com/deepmind/acme
- Have you used any good DRL library?
- Is there a way to get PPO controlled agents to move a little more gracefully?
-
Worthwhile to convert custom env to be dm_env compatible?
Can anyone speak to their experience using acme (https://github.com/deepmind/acme) and by extension dm_env (https://github.com/deepmind/dm_env)? I'm wondering if it would be worthwhile for me to invest the time into converting my custom environment (which loosely follows the standard RL setup) over to this format.
-
[D] Physics and Reinforcement Learning - Discussion of Deepmind's work
acme/acme/agents/tf/mpo at master · deepmind/acme · GitHub
- Applied resources in Pytorch?
-
deepmind acme compatible with windows?
after installing it in a clean env, I tried to run the example provided for solving the gym cartpole env: https://github.com/deepmind/acme/blob/master/examples/control/run_d4pg_gym.py
-
Spec for RL agent implementation?
Acme has a slightly different one: https://github.com/deepmind/acme which includes specs for agents, buffers etc. It is very general. You can see their component description here: https://github.com/deepmind/acme/blob/master/docs/components.md
What are some alternatives?
drqv2 - DrQ-v2: Improved Data-Augmented Reinforcement Learning
dm_env - A Python interface for reinforcement learning environments
tm-dashboard - Dashboard for Trackmania displaying a bunch of vehicle information on screen.
Mava - 🦁 A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX
wandb - 🔥 A tool for visualizing and tracking your machine learning experiments. This repo contains the CLI and Python API.
dm_control - Google DeepMind's software stack for physics-based simulation and Reinforcement Learning environments, using MuJoCo.
softlearning - Softlearning is a reinforcement learning framework for training maximum entropy policies in continuous domains. Includes the official implementation of the Soft Actor-Critic algorithm.
MPO - Pytorch implementation of "Maximum a Posteriori Policy Optimization" with Retrace for Discrete gym environments
vgamepad - Virtual XBox360 and DualShock4 gamepads in python
tonic - Tonic RL library
alf - Agent Learning Framework https://alf.readthedocs.io
selfhosted-apps-docker - Guide by Example