Robot learning teams usually collect demonstrations in two ways: egocentric human video and robot teleoperation. Both produce training signal. They solve different bottlenecks.
Egocentric collection
Head or wrist cameras capture first-person hand and object contact in real environments. Hardware is lighter than a full teleop cell, so coverage across kitchens, shops, and light industrial spaces scales faster. The tradeoff is embodiment transfer: policies must bridge human hands to your robot.
Teleoperation
A human drives the robot. Actions land in the robot’s joint or end effector space, which simplifies early imitation learning. The tradeoff is cost and speed: cells, operators, and robot hours limit how fast you grow task diversity.
How to choose
Use teleop when embodiment match and action labels must be exact. Use egocentric when you need scale, natural contact, and in-the-wild scenes. Many programs mix both: egocentric for breadth, teleop for embodiment-specific fine-tuning. ROBOTRAIN runs egocentric collection and can include teleop when your brief requires it.