Most robots are programmed in one of two ways: by learning from demonstration, in which they watch a task being done and copy it, or by motion-planning methods, in which a programmer spells out the task's goals and limits. A robot that learns a skill by watching often cannot easily use that skill in a new situation and stay accurate. In 2017 MIT scientists built a system called C-LEARN that lets people who cannot code teach a robot a task by giving it knowledge about how objects are handled and then showing it one demonstration. In 2026 University of Maryland researchers reported a system called HumanEgo that can teach robots new hand skills from as little as 30 minutes of first-person video of people doing the tasks.

A big challenge in teaching robots from human video is the embodiment gap: human bodies and robots look, move and perceive the world differently, so a system like HumanEgo focuses on the essence of a task instead of imitating human movements. As of November 2026, a University of Maryland researcher says many robot learning systems still rely on hundreds or thousands of demonstrations from the robot itself. A UC Berkeley robotics researcher says robots face three challenges in grabbing an object: perception, which is seeing where it is, control, which is moving the hand accurately, and physics, which is how the object may shift when touched.

Industrial robots do well at handling specific, pre-defined objects over and over, but handling objects outside such tightly controlled settings is one of the greatest challenges in robotics. Perception is still hard for robots: they can get confused by shiny or transparent objects and have a hard time recognizing everyday objects. A robot that carries its own power faces a trade-off: a larger battery gives more power but also makes the robot heavier, which takes more energy to move. Picking things up is hard for robots, and a robot has trouble picking up an actual chess piece, even though a computer can beat a chess grandmaster. Moravec's paradox describes the pattern: reasoning tasks that seem hard to people are fairly easy to program into computers, while tasks that come easily to us, like moving around, seeing or grasping things, are quite hard to program.

The Mars rover Perseverance can drive itself: its AutoNav system makes 3D maps of the ground ahead, spots hazards and plans a route around obstacles without waiting for directions from Earth. It has a feature called thinking while driving, which processes pictures of the ground while its wheels are turning. Perseverance might reach a top speed of 393 feet an hour with AutoNav, compared with about 66 feet an hour for Curiosity, an earlier Mars rover. In 2023 it crossed a boulder field called Snowdrift Peak in only six autonomous drive sols, or Martian days, which a rover planner said was probably 12 sols faster than Curiosity would have taken. Even so, Perseverance cannot get by on AutoNav alone, and the rover team on Earth is still part of its driving. Ingenuity, a small robot helicopter that flew on Mars, flew by itself, without a person steering it.

Robotic surgery does not replace the surgeon: the surgeon controls the robot arm. The U.S. Food and Drug Administration says a robotic surgery device is not actually a robot, because it cannot perform surgery without direct human control. One cleared surgical system lets the surgeon sit at a console with a 3-D high-definition view and control three robotic arms remotely. One surgical robot system has force feedback, which helps the surgeon feel how stiff the tissue is that the robot arm is holding. Robotic surgery can take longer than other methods because of the time needed to set up the robot. As of November 2026, the U.S. Food and Drug Administration has cleared robotic surgery devices for trained doctors to use in an operating room for several kinds of keyhole surgery, but it has not approved any for preventing or treating cancer.

A U.S. government standards lab, the National Institute of Standards and Technology, or NIST, says robots still have trouble coordinating, communicating and understanding what the other robots and people around them are doing, and that makes them less useful when a task needs more than one robot. As of March 2025, NIST has said there is a complete lack of ways to measure how well robots work with other robots and with people. NIST develops test methods and measurements for robots that work together with people and other robots, so that they finish tasks safely and correctly.

Many robot accidents at work happen when people are programming, fixing, testing, setting up or adjusting the robot. A worker who steps inside a factory robot's working envelope can be hurt if the robot makes an unintended move. The U.S. Occupational Safety and Health Administration says that, as of November 2026, it has no standards written specifically for the robotics industry. Workers must know the hazards when working with robots, because it warns that being unaware of them can be fatal. One of the most important jobs robots do is protect people from injury by doing dangerous jobs in their place. Real robots are made mainly to help people, not to be movie monsters.

Words to know

• embodiment: having a body
• perception: seeing and sensing
• demonstration: showing a task
• autonomous: on its own
• paradox: puzzling idea