UE Autonomous robotics

Degrees incorporating this pedagocial element :


An autonomous system is an artificial system (whether real or virtual) having perception and action capabilities designed to perform a given set of tasks. Autonomous Navigation in a given environment is such a task. This task consists in moving safely (i.e. while avoiding collisions) in this environment in order to reach a given goal, to follow a given route, or to explore/observe a given area. Performing such a task requires to combine three main software components, aimed respectively at (1) constructing a model of the environment, (2) deciding of the actions to be executed by the robot for reaching safely the given goal, and (3) determining the motion commands to be applied to the robot controller for executing the required displacements. An important material exists in the literature to solve this problem in simplified academic cases. However, navigation in open and dynamic environments (i.e. partially known environments featuring moving objects and where time and dynamics have a major importance) remains an open problem.  
The purpose of this course is to present the tools and techniques which are required to achieve autonomous navigation in open and dynamic environments. The first part of the course is devoted to the analysis of the problem, to the presentation of the main problem components and of the related approaches, and to a review of the state of the art. Then, the course will successively address the related world modeling, control, and decision-making issues.
Modeling the real world is mainly done using different types of sensors (e.g. vision, laser, radar…) providing raw and noisy data about the sensed world. To be useful, this data needs to be interpreted in order to build a model that can be used for the decision-making and plan execution functions. The objective of the second part of the course is to present and discuss the main existing approaches and representations which are used for interpreting sensor data and for building appropriate models of the world. This problem is known as the “mapping problem”.  Since the world is dynamic, it is also necessary to distinguish and identify the static and the dynamic parts of the world. The static part is used to localize the autonomous robot, i.e. to find its position/orientation in the world reference frame. The dynamic part (i.e. the moving objects present in the robot environment, along with their respective positions/orientations and velocities) is clearly a key information for being able to make collision-free navigation decisions. These two problems are respectively known as the SLAM (Simultaneous Localization and Mapping) and the DATMO (Detection and Tracking of Mobile Objects) problems.
Once a model of the world is available, the system exploits this model in order to determine the most appropriate motion actions to be applied to carry out the navigation part. The third and fourth parts of the course are respectively devoted to the presentation of the control and of the safe navigation issues. The part concerning the control successively addresses the involved control approaches and the underlying mechanical and mathematical algorithms. The navigation part mainly addresses the problem known as “Motion Planning”, i.e. the a priori computation of a collision free motion toward a given goal. In a first step, the basic models and approaches will be presented; then the way to expand theses basic models for handling the different motion constraints imposed by the autonomous system itself (kinematics, dynamics) or by its environment (collision avoidance for dynamic obstacles, uncertainty) will be addressed.