Artificial Intelligence and Robotics

The course is organized in 4 modules of 3 credits. Each module has its own specific program available at the beginning of the lesson cycle. A detailed updated description of the programs is available at the page: http://www.diag.uniroma1.it/vendittelli/EIR/ Here is a summary: Module 1: Modeling and control of multi-rotor UAVs (Marilena Vendittelli) Introduction to the course and aerial vehicles modeling Quadrotor modeling. Control based on linear approximation. Backstepping-based control. Control based on dynamic feedback linearization. Geometric control on SE(3). State estimation. Motion planning. Controllers comparison. Fault-diagnosis and fault tolerant control - part I: introduction. Fault-diagnosis and fault tolerant control - part II: application to quadrotors. Module 2: Underactuated Robots (Leonardo Lanari, Giuseppe Oriolo) Introduction Motivation. Definition of underactuated system (generalized coordinates vs degrees of freedom). Examples of underactuated robots. Modeling and Properties Eulero-Lagrange modeling (classic and alternate). State-space form. Control problems of interest. Controllabiity (STLA, STLC, natural controllability). Comparison with fully actuated robots. Integrability conditions for passive dynamics. Equilibrium points and linear controllability. Case Studies: Acrobot and Pendubot Modeling. Approximate linearization at equilibria. Linear controllability. Balancing. Partial feedback linearization. Swing-up (1) via analysis of the zero dynamics (2) via energy pumping. Zero dynamics in underactuated systems Normal form and zero dynamics. Importance of the zero dynamics in control. Zero-dynamics in linear and nonlinear underactuated systems. The homoclinic orbit. Passivity Definition and physical interpretation. Linear and nonlinear mechanical systems examples. Dissipativity in state space representations. Feedback equivalence to a passive system. Output stabilization of passive systems Energy-based control of underactuated systems The convey-crane and reaction-wheel cases. Optimization methods for Planning and Control Introduction to Dynamic Programming. Hamilton-Jacobi-Bellman equation. Derivation of the Linear Quadratic Regulator Linear-Time-Varying LQR. Trajectory optimization with Iterative LQR. Constrained optimization. Model Predictive Control (Linear, LTV and Nonlinear). LQR-trees. Module 3: Locomotion and haptic interfaces for VR exploration (Alessandro De Luca) General introduction to haptic and locomotion interfaces with several illustrative examples. Two specific hardware devices: the Geomatic Touch haptic interface; the Cyberith Virtualizer locomotion interface (with the Oculus Rift HMD). Possible applications of these interfaces. Design, construction, actuation, sensing, modelling, and system issues for two locomotion platforms for VR exploration, developed within the CyberWalk project: the CyberCarpet (ball-array) and the CyberWalk platform (2D, omni-directional). Control design and experimental validation for the CyberCarpet. Control design and experimental validation for a 1D treadmill. Control design, experimental validation, and perceptual evaluation for the 2D CyberWalk platform. Module 4: Control of Multi-Robot Systems (Andrea Cristofaro) Examples of applications of multi-robot systems. Centralized vs. decentralized architectures. Elements of graph theory. Connectivity and Consensus; Passivity and Lyapunov stability; Interconnection of mechanical systems. Application to multi-UAV systems: Formation control with time-varying topology; Formation control with connectivity maintenance; Steady-state behaviors; Overview of other multi robot problems.

A general background in robotics (kinematics, dynamics, planning, control) is desirable but not mandatory.

Material distributed by the lecturers.

Lectures illustrating the methodologies used in the areas considered in the different modules of the course. Case study analysis, examples of application to real systems. Hands-on laboratory activities, if allowed by the anti-covid measures.

Not mandatory in general, although instructors of each module may overrule this general modality.

Projects or presentations for topics in each module.

Lectures illustrating the methodologies used in the areas considered in the different modules of the course. Case study analysis, examples of application to real systems. Hands-on laboratory activities, if allowed by the anti-covid measures.