← Back to Portfolio

Attention Heat Map Visualization for VLA Models

Visualizing attention maps for VLA models to understand how the model focuses on different parts of the input.

Overview

This project demonstrates how robotic arms can learn using Visual Language Action (VLA) models like SmolVLA & GR00T N1.5 and imitation learning like Action Chunking Transformer(ACT) .

Project Details

This project demonstrates how robotic arms can learn using Visual Language Action (VLA) models and imitation learning.

Technical Details

The system uses ROS2 for robot control and PyTorch for model inference.

  • Framework: ROS2 Humble
  • Model: Action Chunking Transformer
  • Inference Speed: 10Hz real-time

Links