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

Min Decision Transformer

Min Decision Transformer is a minimal Python and PyTorch implementation of the Decision Transformer method for offline reinforcement learning on MuJoCo control tasks in OpenAI Gym. It uses a causal GPT-style transformer and provides scripts for training, testing checkpoints, rendering evaluation episodes and plotting logged data, alongside a Google Colab notebook for training and visualization.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Min Decision Transformer — 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 to study or run the repository’s offline reinforcement-learning workflow when you have formatted D4RL data and the required MuJoCo dependencies, or to explore the provided Colab notebook.

Who it is for

  • Researchers exploring Decision Transformer and sequence-modeling approaches to offline reinforcement learning.
  • Developers seeking a compact PyTorch baseline for MuJoCo control tasks.
  • Learners who want to follow a notebook-based training and visualization workflow.

Documented uses

  • Train a Decision Transformer on formatted D4RL data for a MuJoCo environment.
  • Evaluate a saved model checkpoint and optionally render an episode.
  • Plot experiment values from logged CSV files.
  • Inspect training and rendered results through the provided Colab notebook.
Before you choose

Requirements & limits

The repository is public and lists an MIT license. It also provides a Google Colab notebook for training and visualization.

  • Training examples specify an environment, dataset and device; the documented HalfCheetah example uses CUDA.
  • Evaluation requires a checkpoint and the same dataset selection used in training to load corresponding state-normalization statistics.
  • The README reports its results after 20,000 updates, compared with 100,000 for the official results, and says the figures are not directly comparable.
  • After formatted D4RL data is saved, the README says the D4RL library is not needed further for training.

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

Train and evaluate a MuJoCo experiment

Use the documented scripts to train a HalfCheetah model and evaluate a checkpoint.

Before you start

  • The repository’s Python scripts
  • Formatted D4RL data in the data directory
  • Installed MuJoCo dependencies
  • A CUDA device for the documented training example, or a CPU device for its testing example
  • A trained checkpoint for evaluation

Steps

  1. Train with the command below.

    Documented command
    python3 scripts/train.py --env halfcheetah --dataset medium --device cuda

  2. Evaluate a checkpoint with the command below.

    Documented command
    python3 scripts/test.py --env halfcheetah --dataset medium --device cpu --num_eval_ep 1 --chk_pt_name dt_halfcheetah-medium-v2_model_22-02-13-09-03-10_best.pt

  3. Add --render to the test command to render the evaluation episode.

Expected outcome

The training script runs the specified experiment; the testing script evaluates a checkpoint and can render an episode.

Guide limitations

  • The example checkpoint name is specific to the README’s example and must correspond to an available checkpoint.
  • Testing requires the dataset used for training so the matching state-normalization statistics are loaded.
  • The README says evaluation uses v3 control environments in mujoco-py.
  • Commands are documented examples, not execution-tested instructions.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
nikhilbarhate99/min-decision-transformer · Official homepage title · retained in its source language
How we review sourcesSuggest a correction