Robot Artificial Intelligence

Description

This course is an introduction to embodied artificial intelligence, where a machine senses its environment, reasons under uncertainty, acts and adapts to change. The course focuses on robot AI: we will explore means of creating artificial robot behaviors where the robot learns about objects, physics, and the world. The course is however designed to teach skills that students will easily generalize to other application domains. Applying AI techniques to a physical system poses challenges that are not apparent in other contexts. This course aims to teach how one casts an embodied-agent problem to a form that lends itself to an AI solution, for instance by choosing AI techniques, sensor/data representations and motor command schemes that are synergetic with one another. This course will cover both "classical" AI techniques that are easily parametrized by an expert, and techniques that are learned from data. We will study the applicability of both approaches and discuss how to judiciously choose one or the other based on the nature of the task. Data-driven models used in production today are generally trained offline, on datasets that have been carefully annotated with the help of experts. It is however possible to let machines learn in a more natural way, by observing the effect of their actions or the actions of others onto the world, i.e., embodied learning . We will study machine learning techniques that are relevant to embodied learning, and practice on real-world problems. At the end of the course, students will be able to: Identify problems that lend themselves to a robot AI solution and decide whether a “classical” or data-driven solution should be preferred. Cast an embodied-agent problem to a form that lends itself to an AI solution. Generate an intelligent robot behavior: Extract information from sensor streams (e.g., object/people identity/position, body postures, 3D room and object structures) and control robot actuators, Learn useful sensorimotor behaviors (e.g., mobility or grasping) Apply their work onto a robot platform.