README.md

metadata

language: en
tags:
  - decision-transformer
  - reinforcement-learning
  - hierarchical-learning
  - hopper
license: mit

Hierarchical Decision Transformer for Hopper

This model extends the Decision Transformer architecture with hierarchical clustering capabilities for improved long-horizon task performance on the Hopper-v3 environment.

Model Description

• Model Type: Hierarchical Decision Transformer
• Training Environment: Hopper-v3
• Input: State observations (11 dimensions), desired returns
• Output: Actions (3 dimensions)
• Architecture Features:
  • Hierarchical clustering head for subtask identification
  • Subgoal-based weighting
  • Multi-task capabilities

Usage

from transformers import DecisionTransformerModel
model = DecisionTransformerModel.from_pretrained("anna4142/hierarchical-decision-transformer")