OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Handful Of Trials

Handful Of Trials is a Python research-code repository for experimenting with model-based deep reinforcement learning using probabilistic dynamics models. It implements probabilistic ensembles with trajectory sampling (PETS) and provides scripts for training and evaluating control tasks in Cartpole, Reacher, Pusher and HalfCheetah.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Handful Of Trials — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you need research code for running or adapting PETS-style model-based reinforcement-learning experiments in the listed simulated control tasks.

Who it is for

  • Researchers studying model-based deep reinforcement learning and probabilistic dynamics models.
  • Developers who want to inspect or modify PETS experiment code for simulated control tasks.

Documented uses

  • Run PETS experiments in the repository’s supported control environments.
  • Adjust controller choices and experiment configuration values.
  • Save and inspect observations, actions, rewards and returns from trials.
  • Render a rollout from saved model files.
Before you choose

Requirements & limits

The repository is public and lists an MIT license. The documented setup requires MuJoCo 1.31 for the provided environments and dependencies from requirements.txt; the README also describes Docker-based setup.

  • The supplied experiment environments are Cartpole, Reacher, Pusher and HalfCheetah.
  • Controller options include deterministic or probabilistic models, ensemble choices, propagation methods and Random or CEM action-sequence optimization.
  • The Docker instructions require a MuJoCo key at /root/.mujoco/mjkey.txt and nvidia-docker for GPU access.
  • The README describes results in logs.mat and requires model.mat and model.nns in the model directory for rendering.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run an experiment in a documented environment

Start a PETS experiment and save its trial results for one of the repository’s listed control environments.

Before you start

  • Python and the repository dependencies installed from requirements.txt.
  • MuJoCo 1.31 for the provided environments; MuJoCo use requires a key.

Steps

  1. Choose cartpole, reacher, pusher or halfcheetah as the environment.

  2. Run the command below.

    Documented command
    python scripts/mbexp.py -env

  3. Optionally set a results directory with -logdir .

Expected outcome

The experiment writes results beneath the selected log directory, including trial observations, actions, rewards and returns in logs.mat.

Guide limitations

  • The repository documentation identifies MuJoCo 1.31 as a requirement for the provided environments.
  • The README notes that using MuJoCo requires a key.
Documentation 1 ↗
Source-based practical guide

Render a saved model rollout

Visualize a rollout using model files produced or saved for the repository’s rendering script.

Before you start

  • A saved model directory containing model.mat and model.nns.

Steps

  1. Locate the directory containing model.mat and model.nns.

  2. Run the command below with the -model_dir argument and the relevant environment options.

    Documented command
    python scripts/render.py

Expected outcome

The rendering script visualizes a rollout with the trained model.

Guide limitations

  • The model directory must contain both model.mat and model.nns.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
kchua/handful-of-trials · Official homepage title · retained in its source language
How we review sourcesSuggest a correction