OVERVIEW

Artificial Intelligence for Robotics II (AI4Ro2) describes advanced concepts. methods and techniques at the intersection between Artificial Intelligence (AI) and Robotics practice. The aim of the course is to provide future scholars and professionals with conceptual tools and practical knowledge about how to integrated cutting-edge AI techniques in robot software architectures, and to guide future engineers about how to do so. The course on real-world, advanced scenarios, e.g., autonomous vehicles, collaborative robots, assistive robots.

AIMS AND CONTENT

LEARNING OUTCOMES

Artificial Intelligence for Robotics 2 is the logic follow-up of Artificial Intelligence for Robotics 1. In this course, the students will be introduced to concepts related to knowledge representation and reasoning (ontologies, description logics, OWL, subsumption, instance checking), planning for hybrid domains (with a particular focus on discrete/continuous domains), as well as AI-based robot motion algorithms (es., RRTs, probabilistic roadmaps, belief-space planning).

AIMS AND LEARNING OUTCOMES

The main objective of Artificial Intelligence for Robotics II (AI4Ro2) is to provide students and scholars with methodological approaches and pragmatic knowledge about how to integrate advanced AI techniques in robot architectures to make them capable of operating in real-world environments in a robust way.

As robots are deployed in scenarios progressively more unstructured, and in conditions whereby their behaviour cannot be easily foreseen in advance, the degree of intelligence they must be provided with becomes of critical importance. Encoding advanced AI algorithms in robots operating in such conditions requires a careful trade-off between their capabilities and the associated computational requirements. The problem of defining an AI-based robot architecture requires:

1. identifying the proper representation level of sensory information, and its use to guide robot behaviour;
2. selecting the most adequate representation approaches to combine sensory information and robot knowledge, which take into account efficiency, computational load, and usability;
3. integrating algorithms able to operate on such represented information according to various reasoning approaches, e.g., induction, deduction, abduction;
4. connecting the reasoning layer with action-oriented robot motion strategies, which are robust to unforeseen sensory data and events.

AI4Ro2 will provide a reasoned treatment of current, state‐of‐the‐art AI-backed perception, cognition, knowledge representation, reasoning, and action approaches, as well as a critical discussion about typical scenarios, use cases, and solutions.

PREREQUISITES

• Robotics Engineering course: Artificial Intelligence for Robotics I
• Advanced knowledge of C/C++.
• Basic knowledge of Java or Python may be helpful.

TEACHING METHODS

AI4Ro2 is organized in four key topics. Each topic will be taught via theoretical concepts followed by practical work. AI4Ro2 will use the flipped classroom teaching methodology. Students are strongly encouraged to propose novel solutions to specific practical problems, which originate from real-world research challenges or industrial needs.

SYLLABUS/CONTENT

AI4Ro2 is a new, highly experimental course. AI4Ro2 mixes up theoretical insights about AI techniques with practical knowledge about how to make them work in robots. AI4Ro2 is organised around the following topics:

TOPIC 1: Introduction and motivations:

1. Introduction to the course
2. Differences and extensions with respect to AI4Ro1

TOPIC 2: Knowledge representation and reasoning: 

1. Knowledge bases and ontologies, the Ontology Web Language (OWL)
2. Description Logic and its extension
3. Reasoning in ontologies: subsumption, instance checking, rules.

Practice classes focused on TOPIC 2.

TOPIC 3: Planning in discrete/continuous domains: 

1. Recap on STRIPS-based planning
2. Using classical planners for tasks with continuous operators in robot tasks
3. Combined task and motion planning
4. Introduction to PDDL+: syntax and semantics
5. Semantic attachments

Practice classes focused on TOPIC 3.

TOPIC 4: AI-based robot perception and motion algorithms:

1. Probabilistic and quantum-like robot perception models
2. Probabilistic motion models
3. Belief-space planning

Practice classes focused on TOPIC 4.

RECOMMENDED READING/BIBLIOGRAPHY

Relevant material will be given by the teachers and the instructors during the lectures.

TEACHERS AND EXAM BOARD

Exam Board

FULVIO MASTROGIOVANNI (President)

ARMANDO TACCHELLA (President Substitute)

LESSONS

LESSONS START

https://corsi.unige.it/10635/p/studenti-orario

Class schedule

The timetable for this course is available here: Portale EasyAcademy

EXAMS

EXAM DESCRIPTION

The AI4Ro2 final mark is based on assignments (50%) and on a written exam (50%). Assignments work as follows:

• at the beginning of October, a number of assignments are proposed;
• students, self-organized in groups of up to 10 people, bid on assignments; each group is required to express 3 ordered preferences;
• we'll try to satisfy the preferences at best and allocate assignments to groups accordingly;
• at that point, work on assignments can start.

Please note that:

• EMARO+ and JEMARO students have a strict deadline to complete their assignment, which is usually the end of February;
• Ph.D. students attending the course can propose a topic on their own, agreed with us, as an assignment, and do not have any specific deadline.

Properly carrying out an assignment means providing:

• an appropriate sketch of the solution the group will aim at designing and developing;
• a (possibly working, maybe with limitations) solution to the given problem;
• properly commented code (in a specific format) and, where appropriate, a tutorial;
• a video showing how the developed solution works.

The written exam will check single students' knowledge about the topics discussed throughout the course. 

ASSESSMENT METHODS

The overall grade will be determined on the basis of:

• the exhibited group's capability in applying the notions and insights discussed during the classes in the assignments;
• the "quality" of the provided documentation for the assignment;
• the outcome of a discussion about the assignment;
• for each student, the results of the written exam.

Exam schedule

FURTHER INFORMATION

Follow-ups:

• for successful assignments, we typically encourage students to co-author a scientific paper for Robotics-related conferences;
• for EMARO+/JEMARO students, a few assignments can be continued as possible MSc thesis topics.