rl
rl copied to clipboard
pytorch
A modular, primitive-first, python-first PyTorch library for Reinforcement Learning.
## Motivation Plan for the doc revamp - [x] A 0-to-1 tutorial or Getting started #861 - [x] A tutorial on building a custom env #911 - [ ] A...
Hi there, As mentioned [here](https://github.com/NVIDIA-Omniverse/Orbit/discussions/108#discussion-5511668), I'm trying to use torchrl with [NVIDIA Orbit](https://github.com/NVIDIA-Omniverse/Orbit) to train an agent in parallel robot environments. I tried to draw inspiration from your recent IsaacGymEnv...
## To Reproduce ```python from torchrl.envs import EnvCreator, ParallelEnv from torchrl.envs.libs.gym import GymEnv def run(from_pixels): env = ParallelEnv( 2, EnvCreator(lambda: GymEnv("CartPole-v1", from_pixels=from_pixels)) ) print(env.reset()) env.close() if __name__ == "__main__": #...
Aren't operations like this blocking? causing performance issues due to GPU synchronization? https://github.com/pytorch/rl/blob/5b3dd988ad7f0f91a2f61f77a8b3380eae3131fd/torchrl/objectives/ppo.py#L443
## Description Added optional `pred-continue` function like https://github.com/danijar/dreamer/blob/master/dreamer.py#L187C32-L187C32, and using `collector-mask` to remove empty data. ## Motivation and Context To make sure dreamer can work on early-stop env. - [...
## Motivation Add A3C implementation as presented [here](https://arxiv.org/abs/1602.01783) so that the experiments of [Rainbow](https://arxiv.org/pdf/1710.02298.pdf) can be added to the examples in further contributions. ## Solution Add A3C code like [A2C](torchrl/objectives/a2c.py)...
## Describe the bug When I run the dreamer example with the default configuration (DMControl Cheetah Run), the program terminates after some time with the following error: ` RuntimeError: Env...
## Motivation The use of a global threading.RLock in replay buffer prototypes limits the number of concurrent reads that can be performed on a replay buffer unnecessarily when contents are...
## Motivation I would propose implementing a simple Transform that remembers the previous action taken. This can be very useful when considering environments with time delay, or when training agents...
