VLA Workshop

Following our previous VLA session, RoboSoc UCL presents VLA Workshop 2, focusing on end-to-end evaluation and execution of Vision-Language-Action policies in simulation.

🗓 Sunday 8 February
⏰ 14:00 – 16:00
📍 Bedford Way (20), IOE – Room W4.01

What to Expect

Part 1: End-to-End VLA Evaluation Workflow

• How VLA policies are evaluated through closed-loop rollouts in simulation
• Understanding the full pipeline:
  policy → simulator → observations → actions → metrics

Part 2: Hands-On Setup

• Set up the full environment, including dependencies, simulator, and GPU/headless rendering
• Load a pretrained VLA checkpoint and run it inside a robotics simulation

Part 3: Observations, Actions & Results

• Stream live observations and run policy inference
• Apply actions correctly in simulation
• Visualise rollouts and interpret performance metrics (e.g. success rate)

This workshop is ideal for students interested in robotics, embodied AI, reinforcement learning, and vision-language models. Some familiarity with Python and basic ML concepts is helpful, but not required.

🚀 Come explore how language, perception, and action come together in modern robotics systems.