Learning To Play Settlers of Catan With Deep RL

In recent years RL has attracted a lot of attention due to the extremely impressive results that have been achieved in games such as Go, DOTA 2and Starcraft 2. Whilst these are all incredible bits of work that demonstrate the amazing potential RL has to offer, it's important to realize that they all required huge teams of world class researchers and software engineers along with tens of millions of dollars worth of compute power. Attempting to use the same techniques to solve anything of comparable complexity is simply not possible at the moment if you have limited compute availability. What this generally means is that actually a majority of research papers in RL use the same set of relatively simple benchmark environments, which whilst completely understandable can get a bit boring. So as much as I am interested in fundamental RL research, I also really like looking at new environments and seeing how much progress it's possible to make using existing RL techniques. As a way to teach

Learning To Play Settlers of Catan With Deep RL Learning to Play Settlers of Catan with Deep Reinforcement Learning Screenshot of the Settlers-RL simulator built for this project. Introduction In recent years RL has attracted a lot of attention due to the extremely impressive results that have been achieved in games such as Go "AlphaGo Zero" - DeepMind (2017) , DOTA 2 "OpenAI Five" - OpenAI (2016-2019) and Starcraft 2 "AlphaStar" - DeepMind (2019) . Whilst these are all incredible bits of work that demonstrate the amazing potential RL has to offer, it's important to realize that they all requi

saved by

• HudZah

related reading

• Part 1: Key Concepts in RL - Spinning Up documentationspinningup.openai.com
• The 37 Implementation Details of Proximal Policy Optimization · The ICLR Blog Trackiclr-blog-track.github.io
• A vision researcher’s guide to some RL stuff: PPO & GRPO - Yuge (Jimmy) Shiyugeten.github.io
• State of RL for reasoning LLMs | A. Weersaweers.de
• Key Papers in Deep RL - Spinning Up documentationspinningup.openai.com
• Andrej Karpathy — AGI is still a decade awaydwarkesh.com
• Learning Beyond Gradientstrinkle23897.github.io
• Algorithms — Ray 2.55.1docs.ray.io
• How DeepMind's Generally Capable Agents Were Trained — LessWronglesswrong.com
• IsoCompute Playbook: Optimally Scaling Sampling Compute for RL Training of LLMscompute-optimal-rl-llm-scaling.github.io
• Deep Reinforcement Learning Doesn't Work Yetalexirpan.com
• HANABI – np – ( ´ ▽ ` )ﾉnphard.io