Research Topics


Behavioral control of aerial manipulators

A three-layers coordinated control for multiple aerial manipulators has been developed and experimentally validated. The first layer is centralized, it communicates with every vehicles and generates the desired trajectories of each end-effector; the second layer, local to each vehicle, computes the motion references; the last layer is a motion controller ensuring the tracking of the previous layer outputs. At the second layer, the overall mission is hierarchically decomposed in a set of basic sub-tasks, called elementary behaviors, which are combined together, in a prioritized way, into more complex tasks, called compound behaviors, by exploiting the Null Space-based Behavioral control paradigm. The compound behavior to be assigned to each aerial manipulator is selected via a local supervisor, designed via a Finite State Automata. Experiments have been conducted on a test-bed composed of two multi-rotor aerial platforms equipped with a 6 DOFs manipulator, developed by CATEC (Centro Avanzado de Tecnologias Aeroespaciales) within the EU-funded project ARCAS (Aerial Robotics Cooperative Assembly System), aimed at developing cooperative free-flying robot system for assembly and structure construction.  


Control of aerial vehicles

Unmanned Aerial Vehicles (UAVs) are being widely used in a number of applications, mostly military but also civilian, involving surveillance of indoor or outdoor environments, remote inspection and monitoring of hostile environments. Among UAVs, quadrotor helicopters are emerging as a popular platform, due to their larger payload capability and their higher manoeuvrability with respect to single-rotor vehicles. Motion control of quadrotors is a widely investigated but still challenging issue, since the quadrotor is an underactuated system and, often, is equipped with limited sensing devices. The research activity related to UAVs has been focoused on the problem of motion control of the end effector of a robot manipulator mounted on a quadrotor helicopter. A hierarchical control architecture has been proposed: in the top layer, an inverse kinematics algorithm computes the motion references for the actuated variables, i.e., position and yaw angle of the quadrotor vehicle and joint variables for the manipulator; in the bottom layer, a motion control algorithm is in charge of tracking the motion references. The proposed motion controller in the bottom layer includes a vehicle position controller computing the thrust force and the reference values for pitch and roll angles, an attitude controller, that, on the basis of these references, computes the moments acting on the quadrotor, and a manipulator controller that computes the joint torques.  


Fault Diagnosis for multi-robot systems

A distributed fault detection, isolation and accommodation (FDIA) strategy for a team of networked robots based on a distributed controller–observer schema has been developed and experimentally tested. The assumption is that each robot computes its own motion control only based on local information and on information from a subset of its teammates, usually named neighbors. Remarkably, different from other works in literature, the proposed approach makes each robot of the team able to detect and isolate faults occurring on all the other robots, even if they are not direct neighbors. By means of a local observer, each robot can estimate the overall state of the team and it can use such an estimate to compute its local control input to achieve global tasks. The same information used by the local observers is also used to compute residual vectors, whose aim is to allow the detection and the isolation of actuator faults occurring on any robot of the team. Adaptive thresholds are derived based on the dynamics of the residual vectors by considering the presence of nonzero initial observer estimation errors, and noise terms affecting state measurement and model dynamics. Once the faulty robot has been identified it is excluded by the team and the other robot reconfigure themselves in order to accomplish the mission. The approach was validated via both numerical simulations and experiments involving a team of five Khepera III mobile robots.  


Control of bimanual robotic systems

Dual-arm/hand object manipulation with multi-fingered mechanical hands is a challenging task. In order to achieve the desired motion of the manipulated object, arms and fingers should operate in a coordinated fashion. In the absence of physical interaction between the fingers and the object, simple motion synchronization shall be ensured. On the other hand, the execution of object grasping or manipulation requires controlling also the interaction forces to ensure grasp stability. From a purely kinematics point of view, an object manipulation task can be assigned in terms of the desired motion of the manipulated object: the map of the desired task into the corresponding joint trajectories of the fingers and arms, requires the solution of an inverse kinematics problem. Internal force control has to be adopted to avoid contact breaking or excessive mechanical stresses on the object. The manipulation system can be redundant also if the single fingers are not: this is due to the presence of the additional degrees of freedom (DOFs) provided by the contact variables. These redundant DOFs can be suitably exploited to satisfy a certain number of additional tasks with lower priority and to fulfill a certain number of constraints during the system's motion. Secondary tasks could bring the system close to violate one or more costraints: a suitable task sequencing algorithm should be considered to detect which tasks are potentially responsible of those violations, disable them and enable again as soon as possible (i.e. when any violation will not occur anymore). Secondary tasks are aimed to improving grasp stability and dexterity while costraints involve system's safety (i.e. joint limit avoidance, collision avoidance between finger,etc.). A two stage control strategy can be pursued: the first stage performs a kinematic inversion with redundancy resolution and task sequencing, while the second provides internal force (or internal force/position) control. The developed control strategy has been tested in the Grasp It! environment. Current research is focused on impedance control of bimanual robotic systems.  


Multi-robot systems

Starting in early 1980's, the attention of researchers was attracted by the idea of creating groups of robots able to collaborate, in order to accomplish one or more predefined tasks. The basic principle behind this new approach to the robot coordination was directly inspired by the observation of natural systems. From an engineering point of view, a Multi-Robot System (MRS) can improve performance, robustness and reliability of a robotic system. The research activity related to this topic has been twofold. First, a layered architecture for the control and mission achievement for MRSs has been designed. The architecture is general in the sense that it can be applied to any kind of mission to be accomplished by a robotic system. As a further contribution, the architecture has been applied to the multi-robot patrolling problem. An effective approach to the control of robotic systems performing complex missions is based on the decomposition of the task in several sub-tasks, commonly termed behaviors. While several paradigms have been used to properly compose commands from different behaviors, they can be mainly classified as competitive, where one behavior at time can be satisfied, cooperative, where the output of multiple behaviors are combined in order to achieve multiple objectives at once, and competitive-cooperative, that presents the advantages of the previous approaches. The Null-Space-based-Behavioral (NSB) approach can be defined as a competitive-cooperative approach, trying to overcome the disadvantages of the above approaches. According to this technique, a secondary task is fulfilled if it does not conflict with an higher level task, while it is released in case of conflict. The standard NSB is a centralized architecture: this might become a limit in the presence of large teams and/or limited communication between. Hence, current research efforts are focused on the development of decentralized control of multi-robot teams.


Motion planning and control of cooperative manipulators

When a cooperative multi-arm system is employed for the manipulation of a common object, it is important to control both the absolute motion of the object and the internal stresses. To this aim, impedance control schemes for cooperative manipulators have been proposed for control of either object/environment interaction forces or internal forces. Namely, when the held object interacts with the environment, large contact forces may arise if the planned trajectory is not consistent with the geometry of the environment. In order to achieve bounded contact forces, an impedance behavior can be enforced between the object's position/orientation displacements and the contact force/moment (external impedance). On the other hand, even when object/environment interaction does not take place, the interaction between the manipulators and the object may lead to internal forces and moments, i.e., mechanical stresses which do not contribute to the object's motion and may cause damage to the system and overloading of the actuators. To counteract building of large values of internal forces, an impedance behavior can be enforced between the position/orientation displacements of each manipulator and the end-effector force/moment, contributing solely to the internal loading of the object (internal impedance). The above described impedance approaches have been combined in a unique control framework, aimed at controlling both the contact forces due to object/environment interaction (external impedance) and the internal forces due to manipulators/object interaction (internal impedance). A task-oriented motion planning approach for general cooperative multi-robot systems is proposed. In order to derive a meaningful task formulation, a taxonomy of cooperative multi-arm systems of industrial interest has been devised. Then, a workpiece-oriented general formulation for cooperative tasks has been devised, where the user is asked to specify the motion of the system only at the workpiece level, while the motion of the single arms in the system is computed via kinematic transformations between the relevant coordinate frames. Based on this task formulation, a new instructions set has been derived to extend classical programming languages for industrial robots to general multi-robot systems.


Fault Diagnosis for nonlinear systems

A control system is often required to ensure not only stability and desired performance during normal operating conditions, but also to guarantee a suitable behavior in the presence of failures and malfunctions. Faults causes are usually classified into internal causes (damage of mechanical parts, actuators or sensors failures) and external causes (sudden environmental changes, disturbances). A Fault Diagnosis (FD) scheme is aimed at detecting the occurrence of faults (fault detection), recognizing their location (fault isolation) and identifying their time evolution (fault identification). The FD techniques developed at the AREA Lab adopt the Analytical Redundancy (AR) paradigm and, in particular: the actual system's behavior is compared to the corresponding expected behavior derived via its mathematical model, in such a way to obtain a set of variables sensitive to the occurrence of faults, i.e., the residuals. Then, the analysis of residuals achieve proper fault detection and isolation. As regards complex systems, mathematical model is only partially known, i.e., it is affected by model uncertainties and measurement noise. Hence, the objective of an efficient fault diagnosis algorithm is twofold: the generation of residuals maximally insensitive to uncertainties and noise and, at the same time, the enhancement of their sensitivity to faults. However, the effects of disturbances and uncertainties are usually not decoupled to those due to faults; thus, suitable trade-off solutions must be devised. To the purpose, state estimation techniques via adaptive diagnostic observers have been developed. The developed FD techniques are applied to chemical reactors, robotic systems and a wastewater treatment plants.


Control of chemical batch reactors

The main goal of a control system for chemical batch reactors is that of imposing a given temperature-time profile inside the reactor vessel, in fact the reactor temperature strongly affects the productivity and the quality of the final products. Of course, temperature control is of the utmost importance to ensure safety of the plant and the human operators, especially in the presence of highly exothermic reactions, where the amount of heat released can become very large and, if the heat generated exceeds the cooling capability, temperature runaway may occur. In industrial practice the temperature can be controlled via the heat exchange between the reactor and a heating/cooling fluid, circulating in a jacket, surrounding the vessel or in a coil inside the vessel. Early approaches to control of chemical processes were mainly based on linear methods, such as PID regulators. Since batch industrial processes can exhibit highly nonlinear behavior and operate within a wide range of conditions, linear controllers must be tuned very conservatively, in order to provide a stable behavior over the entire range of operation, thus leading to a degradation of performance. Hence, in the last two decades, nonlinear model-based control strategies began to be preferred for complex processes, thanks to the development of accurate experimental identification methods for nonlinear models and to significant improvements of computing hardware and software. The research activities related to this topic have been focused on the development of a model-based control scheme for a jacketed batch reactor, based on a combination of an adaptive nonlinear observer and an adaptive cascade control. In detail an adaptive observer is designed to estimate the heat released by the reaction, that cannot be usually measured in real time, where the adaption involves the heat transfer coefficient, usually poorly known in industrial applications. Then, a direct adaptive scheme is designed, based on the closure of two control loops. Namely, an outer control loop, closed on the reactor temperature, computes the reference signal for the inner control loop closed on the jacket temperature. The two-loop arrangement improves the robustness of the scheme, while preserving a simple structure for both the controllers. It must be remarked that the observer and the controller can be designed and tuned separately. The convergence of the observer estimation errors and of the controller tracking errors are proven via a Lyapunov-like analysis.