Here is a fast reinforcement learning algorithm for humanoid robots with an open source code that comes with everything you need to run your own humanoid RL experiments. FastTD3 solves humanoid tasks faster other off-policy RL algorithms. It can solve challenging HumanoidBench tasks with dexterous hands in under 3 hours.
As the researchers explain, they built an algorithm that works without bells and whistles. As they show on their website, this framework can be used for sim-to-real transfer.
[HT]