Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Experiment code for "Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Models"
| Date | Stars |
|---|---|
| 2026-07-31 | 476 |
| 2026-08-01 | 478 |
| 2026-08-02 | 479 |
| 2026-08-03 | 480 |
| 2026-08-06 | 480 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Models
<p align=center>
<img src="img/cartpole.png" width=200><img src="img/reacher.png" width=200><img src="img/pusher.png" width=200><img src="img/halfcheetah.png" width=200>
</p>
<p align="center">
<a href="https://arxiv.org/abs/1805.12114">View on ArXiv</a>
|
<a href="https://sites.google.com/view/drl-in-a-handful-of-trials/home">View website</a>
</p>
**Abstract:** Model-based reinforcement learning (RL) algorithms can attain excellent sample
efficiency, but often lag behind the best model-free algorithms in terms of asymptotic performance,
especially those with high-capacity parametric function approximators, such as deep networks.
In this paper, we study how to bridge this gap, by employing uncertainty-aware dynamics models.
We propose a new algorithm called probabilistic ensembles with trajectory sampling (PETS) that
combines uncertainty-aware deep network dynamics models with sampling-based uncertainty propagation.
Our comparison to state-of-the-art model-based and model-free deep RL algorithms shows that our
approach matches the asymptotic performance of model-free algorithms on several challenging
benchmark tasks, while requiring significantly fewer samples (e.g. 25 and 125 times fewer samples
than Soft Actor Critic and Proximal Policy Optimization respectively on the half-cheetah task).
## Requirements
The provided environments require MuJoCo 1.31. Other dependencies can be installed with `pip install -r requirements.txt`.
Alternatively, we have included a Dockerfile for your convenience; this repository will be found in `/root/handful-of-trials` within the image.
We have also provided a prebuilt image; to use the image, run `docker pull kchua/handful-of-trials`.
A few notes on the provided Docker image:
1. Note that using MuJoCo requires that the user have a key; to use the Docker image, ensure that the key is mounted so that it is located at `/root/.mujoco/mjkey.txt`.
2. This image allows the scripts to access user GPUs; consequently, the image must be launched using `nvidia-docker`.
## Running Experiments
Experiments for a particular environment can be run using:
```
python scripts/mbexp.py
-env ENV (required) The name of the environment. Select from
[cartpole, reacher, pusher, halfcheetah].
-ca CTRL_ARG (optional) The arguments for the controller
(see section below on controller arguments).
-o OVERRIDE (optional) Overrides to default parameters
(see section below on overrides).
-logdir LOGDIR (optional) Directory to which results will be logged (default: ./log)
```
To run experiments with default arguments to reproduce our results, only specify the environment (and optionally a log directory). For example: `python scripts/mbexp.py -env halfcheetah`.
Results will be saved in `<logdir>/<date+time of experiment start>/`.
Trial data will be contained in `logs.mat`, with the following contents:
```
{
"observations": NumPy array of shape
[num_train_iters * nrollouts_per_iter + ninit_rollouts, trial_lengths, obs_dim]
"actions": NumPy array of shape
[num_train_iters * nrollouts_per_iter + ninit_rollouts, trial_lengths, ac_dim]
"rewards": NumPy array of shape
[num_train_iters * nrollouts_per_iter + ninit_rollouts, trial_lengths, 1]
"returns": Numpy array of shape [1, num_train_iters * neval]
}
```
#### Rendering
A script for visualizing a rollout with a trained model is also provided for convenience, and can be run using `python scripts/render.py`. Usage is similar to `mbexp.py`, with an additional argument `-model_dir <path to directory containing saved model files>`. The provided directory must contain `model.mat` and `model.nns`.
### Controller Arguments
To specify controller arguments `-ca`, choose among the following options:
```
model-tExcerpt of 9,176 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:735e21f3541109a9, topic:reinforcement-learning, desc:reinforcement learning