Towards Intelligent Robots

Artificial Intelligence (A.I.) and Robotics, as two main scientific domains related to Computer Science, both have a long history since their introduction. If these two domains were close 40 years ago (e.g., with the robot SHAKEY using the first task planner STRIPS), they seem to have followed separate paths since then, with specific academic forums for each. In this paper, we first review how A.I. is perceived by Robotics researchers (a library of efficient algorithms) and its opposite (an application domain). Facing this reality, we then advocate that a unifying notion is the one of intelligent robots (robots on one side and intelligence on the other), the main characteristics of intelligence for robots being to be able to survive in their environment. We propose the notion of multilayered software architecture of robotic agents, as the key to reach this goal. By abstracting our discussion, we advocate that the key theoretical difference between the two scientific domains lays (1) in the uncertainty inherent to Robotics and not necessarily in A.I., and (2) in the difference between continuity and discreteness (and how to mix them). As an example of intelligent robot, we consider automated terrestrial vehicles (autonomous cars) from the Intelligent Transportation Systems domain (ITS), in which supervision (a Robotics notion) can be considered as an upper layer of their software architectures. Such high level common notion seems more fruitful and useful than merely improving a specific component (i.e., perception, path planning or control) by importing another A.I. algorithm.