Abstraction-MPC Integration (2025 – present)
A framework for decision-making under uncertainty in nonlinear stochastic systems that must meet safety specifications with a guaranteed probability. A novel Interval MDP abstraction maps each abstract action to a set of inputs, so one verified abstract policy yields a whole family of provably safe policies. Model predictive control (on line mixed-integer optimization) then searches that family online to minimize a performance cost, while keeping the safety guarantee. It improves performance by up to 52% over single-policy abstraction methods, with a small loss in the guarantees. The work is under review at Automatica.
Code: github.com/alessandro-riccardi/abstraction-mpc-integration
The European Economic Area Electricity Network Benchmark (2022 – present)
An open-source benchmark of the European electricity grid for testing centralized and distributed control. It covers 27 control areas built from real ENTSO-E and UK DUKES data, resampled to any control time step. It supports centralized and ADMM-based distributed MPC solved with Gurobi, and interchangeable partitioning strategies. It is also the testbed for multi-agent RL work on load frequency control. The work is published as a chapter in Springer’s Control Systems Benchmarks (2025) and at the 2025 European Control Conference (ECC).
Code: github.com/alessandro-riccardi/the-european-economic-area-electricity-network-benchmark
Network Partitioning (2021 – 2025)
Non-centralized control of large-scale systems (decentralized, distributed, hierarchical or coalitional) starts with a partitioning problem: how to decompose a network of dynamical systems into units that local controllers can manage. I developed a general partitioning framework that selects fundamental system units and aggregates them into composite units, either algorithmically or by optimization. It is guided by a global network metric, the partition index, which balances interactions within and between units at a chosen level of granularity. In distributed model predictive control case studies on linear and hybrid systems, it significantly reduced computation time and cost while maintaining or improving control performance. The project also includes a systematic review of partitioning techniques for non-centralized MPC. It contributes a unified graph-theoretical formalism, a mixed-integer programming reformulation, the concepts of predictive partitioning and multi-topological representations, and a methodology for defining quality metrics. The work is published in Annual Reviews in Control and at the IEEE Conference on Decision and Control (CDC).
Code general partitioning strategy: github.com/alessandro-riccardi/a-general-partitioning-strategy-for-non-centralized-control
Code genetic-optimization-based partitioning strategy: github.com/alessandro-riccardi/ga-partitioning
Manipulator Control for Remote Actuation of Soft Robots (2021)
Magnetic fields can actuate miniaturized robots remotely, with promising medical applications. I identified the magnetic field of a mobile electromagnet that can generate a desired field across a large 3D workspace. I then built a control model of the field-generation system to produce a field that drives a locomotion gait in a legged miniaturized robot. Preliminary experiments confirmed the approach. Published in IEEE Robotics and Automation Letters, 2023.
Control of UAVs and Aerial Vehicles (2019 – 2020)
I designed an aerial vehicle for tasks such as load carrying, teleoperation and rescue, in which two independently controlled drones actuate a central rigid body. Each drone is attached to the body by a 2-DoF actuated joint and a spring suspension. The four joint rotations and two thrusts give exactly six control inputs for the body’s six degrees of freedom, so the vehicle is not over-actuated. I modeled the full dynamics, including the joint kinematics, spring forces and torques, and joint dynamics. I then designed a layered controller. A vehicle controller uses the load’s internal force, so the drones coordinate through the forces exchanged via the body, with no explicit communication. A joint attitude planner converts the desired forces into thrusts and joint angles, and a joint controller tracks them. A unicycle-like trajectory planner lets the vehicle reach poses beyond the controller’s region of attraction, which is empirically below 5 m in radius. In simulation, the vehicle regulates its pose, tracks circular trajectories and stays stable under a step change in payload weight. An unbalanced load produced a steady-state angular error, so I proposed a calibration algorithm that adjusts how the weight is shared between the drones while hovering. It restores the performance of the balanced case.
Kinesiometer: A Prototype for Handgrip Strength Measurement (2018 – 2019)
A fully tested, working prototype for measuring handgrip strength, which I developed on my own from the client’s specification to final delivery. I defined the concept, designed the device in 3D, and verified its mechanical strength with CAD strength tests. It was 3D printed following medical material standards. A linear actuator provides the test force, and a resistive pressure sensor measures it in kilograms. The hand-opening force is computed as the peak strength of the handgrip. On the electronics side, I selected the components, designed the circuit schematics, and built the system around an Arduino. I also wrote the software of over 1,000 lines of code that runs the test, shows the measurements on a display, and stores the results so they can be recalled later.