Project detail

SO101 Arm · Visual Language Action Deployment

We are training the SO101 arm to autonomously play tic-tac-toe using GR00T 1.6 + a local VLM on Jetson Thor.

Goal

Demonstrate fully autonomous tic-tac-toe on SO101.

Hardware

SO101 6-DoF arm and Jetson Thor

Focus

VLA inference, control reliability, and onboard training experiments.

Mission

Make SO101 play tic-tac-toe autonomously.

GR00T 1.6 and a lightweight local VLM are our current stack for perception, move selection, and execution. Tic-tac-toe is the primary benchmark before scaling to broader tasks.

  • SO101 follower/leader teleoperation
  • Teleop-to-dataset pipeline
  • GR00T 1.6 finetuning for tic-tac-toe
  • Local VLM integration for task determination
  • Release reproducible deployment guide for campus outreach