Robotics and world model
Learns how the physical world responds to actions, so robots can see, plan and move.
What it is
World models predict what happens next in an environment, and vision-language-action (VLA) models turn camera images and an instruction into robot movements. Together they are how AI is moving from screens into physical machines.
How it works
A VLA starts from a vision-language model and adds an output that predicts motor actions, trained on recordings of robots doing tasks. World models learn to predict future frames or states from current observations and actions, which lets a system rehearse plans before acting.
What it is good for
- Robot arms that follow spoken instructions
- Simulation and planning for autonomous machines
- Generating training data for robots
Limits to know
- Robot data is scarce and expensive
- Behaviour can fail in unfamiliar settings
- Safety around people needs strict limits
