Foundations · Beginner · 20-40 hours
Computer Vision
Methods that let machines extract useful structure from images and video.
Direct answer
What is Computer Vision?
Methods that let machines extract useful structure from images and video.
Definition and scope
Methods that let machines extract useful structure from images and video.
Vision pipelines use learned representations for detection, segmentation, depth, tracking, and 3D understanding.
Why it matters
Robots need perception for objects, geometry, motion, people, and scene context.
How it works
Vision pipelines use learned representations for detection, segmentation, depth, tracking, and 3D understanding.
Beginner learning path
Start with image formation, convolutional networks, detection, segmentation, and camera geometry.
Recommended next topics
Primary sources
Key papers
RT-1: Robotics Transformer for Real-World Control at Scale
RT-1 trains one transformer policy on a large multi-task dataset of real robot demonstrations.
OpenVLA: An Open-Source Vision-Language-Action Model
OpenVLA is an open 7B-parameter VLA trained on the Open X-Embodiment dataset.
Ego4D: Unscripted First-Person Video Dataset
Ego4D is a large first-person video dataset for studying what people see, remember, manipulate, and anticipate.
Ego-Exo4D: First- and Third-Person Skilled Activity Dataset
Ego-Exo4D pairs synchronized first-person and third-person video with language, gaze, audio, pose, and 3D signals.
V-JEPA 2: Self-Supervised Video Models for Physical Planning
V-JEPA 2 learns predictive video representations that support visual understanding and zero-shot robot control.
Research ecosystem
Organizations working in this area
Organization
Meta AI
Embodied perception, video prediction, egocentric AI
View profile →Organization
Google DeepMind
World models, robot learning, VLA systems, embodied reasoning
View profile →Organization
World Labs
Spatial intelligence, multimodal world models, generative 3D environments
View profile →Common questions
Frequently asked questions
What is Computer Vision?
Methods that let machines extract useful structure from images and video.
Why does Computer Vision matter for Physical AI?
Robots need perception for objects, geometry, motion, people, and scene context.
How should a beginner learn Computer Vision?
Start with image formation, convolutional networks, detection, segmentation, and camera geometry.