Many applications here - imagine a robot vacuum cleaner, for instance. But robotics really demonstrates the failures of AI. Mike Brady, the head of the Oxford Robotics Group, declared in a public lecture two years ago that he would consider his life work done if he could design a robot that was as competent as a sheep! Although AI has been quite successful at ``rational thought'' tasks like language analysis and game-playing, it is much less so for tasks that involve perception and action. Although our TV cameras can provide all the information a robot might need, we are nowhere near being able to use it in a general-purpose vision system. Much of robotics involves designing sensors to compensate for the limitations of robotic thought: for example, by projecting grids onto the scene to help locate objects.
Because classical symbolic AI has been so unsuccessful here, many people are trying alternative approaches.