Reinforcement Learning
If you have heard of RL, you have inevitably heard of the wonderful Sutton Barto book, which is an excellent resource to get started with. However, if you're anything I am and get easily bored with books, there is a vast amount of incredibly useful resources that may help you absorb the material much quicker and in much more depth. Here are some that I have found useful:
Lecture series
If you're new to RL, the first ones I would recommend would be:
- David Silver's RL course, and
- Martha & Adam White's RL specialization if you can spare a little more time
Once you have a decent grasp of the basics, I would highly recommend the following courses from the Berkeley and Stanford giants that will surely take you to the state-of-the-art RL:
- Pieter Abbeel's Foundations of Deep RL
- Sergey Levine's Berkeley CS285 course
- Chelsea Finn's Stanford CS330 course
- Deep RL Bootcamp 2017
Blogs & wikis
- RL discord wiki hosts a ton of useful resources with many overlaps. Newcomers may want to check out their debugging advice if nothing else
- Lilian Weng's blog posts are an invaluable resource and service to the community
- OpenAI Spinning up is a practically-oriented tutorial/course
- BAIR blog has monthly RL highlights from Berkeley & around
- RL weekly is a series of weekly (not anymore) RL highlights
- Julien Vitay's blog
- Jonathan Hui's Deep RL Series on Medium
Libraries
RL
- Stable Baselines 3 is one that I have extensively used and prefer
- RLlib is a library for large-scale distributed RL applications
- Clean RL is one I haven't used but heard great things about
- My SB3 vs RLlib github repository benchmarking their performance and some of my notes on working with SB3 & RLlib
Imitation Learning
- Imitation has several SOTA imitation learning algorithms
- Seals is the sister library for imitation with all environments
Environments
- OpenAI gym environments
- PyBullet Gym open-source version of Mujoco
- MuJoCo the now-open-sourced physics engine for all control environments
- MiniGrid minimalistic grid-world environments
- MyoSuite has a suite of tasks for musculoskeletal control
- Unity ML-Agents enables games to serve as environments
- ViZDoom for Reinforcement Learning from Raw Visual Information
- Deepmind Control suite
- OpenSpiel RL search and planning games
- Meta-World for multi-task meta RL
- CARLA open-source simulator for self-driving cars
- PettingZoo multi-agent RL environments
- Awesome RL environment list