Training VLAs for coordinated manipulation and navigation tasks.

About This Project

An Isaac Sim 5.1.0 environment for the Spot robot with arm, designed to train multiple VLAs (Vision-Language-Action models) for separate control of legs and the arm.

README

This project creates a simulation environment for Boston Dynamics' Spot robot with an attached arm, enabling research on multi-agent control policies.

Key Features

  • Isaac Sim 5.1.0 - NVIDIA's robotics simulation platform
  • Spot Robot - Simulation of Boston Dynamics' quadruped robot
  • Multi-Agent Architecture - Separate VLAs for leg and arm control
  • Vision-Language-Action Models - Generalist robotic control

Research Goals

  • Train separate policies for locomotion and manipulation
  • Explore multi-agent coordination in embodied AI
  • Develop VLA-based robot control systems

Technical Details

The environment leverages NVIDIA's Isaac Sim for physics-accurate simulation, enabling development of control policies that can potentially transfer to real hardware.