Fabio Amadio

Fabio Amadio

Auto-generated profile, based on publicly available information

Fabio Amadio

Inria, Université de Lorraine, CNRS

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
12 July 2022 in IEEE Transactions on Robotics
Verified
Journal Article
Closed access
FAFabio Amadio
Fabio Amadio
ALAlberto Dalla Libera
Alberto Dalla Libera
RARiccardo Antonello
Riccardo Antonello
DNDaniel Nikovski
Daniel Nikovski
RCRuggero Carli
Ruggero Carli
DRDiego Romeres
Diego Romeres
In this article, we present a model-based reinforcement learning (MBRL) algorithm named Monte Carlo probabilistic inference for learning control (MC-PILCO). This algorithm relies on Gaussian processes (GPs) to model the system dynamics and on a Monte Carlo approach to estimate the policy gradient. This defines a framework in which we ablate the choice of the components, which are the selection of the cost function, the optimization of policies using dropout, and an improved data efficiency through the use of structured kernels in the GP models. The combination of the aforementioned aspects affects dramatically the performance of MC-PILCO. Numerical comparisons in a simulated cart–pole environment show that MC-PILCO exhibits better data efficiency and control performance w.r.t. state-of-the-art GP-based MBRL algorithms. Finally, we apply MC-PILCO to real systems, considering, in particular, systems with partially measurable states. We discuss the importance of modeling both the measurement system and the state estimators during policy optimization. The effectiveness of the proposed solutions has been tested in simulation and on two real systems, which are the Furuta pendulum and the ball-and-plate rig.

4 July 2022 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
MLMarco Laghi
Marco Laghi
LRLuigi Raiano
Luigi Raiano
FAFabio Amadio
Fabio Amadio
FRFederico Rollo
Federico Rollo
AZAndrea Zunino
Andrea Zunino
Arash AjoudaniArash Ajoudani
Arash Ajoudani
Teleoperation offers the possibility to combine human intelligence with robot power and endurance, making it a perfect solution for hostile-for-human environments. Nonetheless, when it concerns prolonged and repetitive operations, classical teleoperation interfaces that replicate one-to-one human commands can be quite demanding in terms of physical and mental efforts. Shared-autonomy approaches, which foresee the combination of direct teleoperation with autonomous control, can help the operator in overcoming such issues without sacrificing the task execution efficiency. Following this philosophy, in this letter we propose a novel technique to assist the operators in reaching and manipulation of objects with one or two arms, combining local visual perception and operator’s action monitoring. In more details, the reaching intention of an operator towards a target object is detected, which is used to adapt robot trajectories autonomously towards a successful grasping pose. In addition, based on the detected size of a target object, single- or dual-arm coordinated movements are autonomously generated without the need for additional human interventions. The experimental results on thirteen participants confirm the potential of the proposed framework in terms of success rate and operator effort.

16 May 2023 in International Journal of Dynamics and Control
Verified
Journal Article
Open access
PDF file
FAFabio Amadio
Fabio Amadio
JDJuan Antonio Delgado-Guerrero
Juan Antonio Delgado-Guerrero
ACAdriá Colomé
Adriá Colomé
CTCarme Torras
Carme Torras
Over the last years, significant advances have been made in robotic manipulation, but still, the handling of non-rigid objects, such as cloth garments, is an open problem. Physical interaction with non-rigid objects is uncertain and complex to model. Thus, extracting useful information from sample data can considerably improve modeling performance. However, the training of such models is a challenging task due to the high-dimensionality of the state representation. In this paper, we propose Controlled Gaussian Process Dynamical Models (CGPDMs) for learning high-dimensional, nonlinear dynamics by embedding them in a low-dimensional manifold. A CGPDM is constituted by a low-dimensional latent space, with an associated dynamics where external control variables can act and a mapping to the observation space. The parameters of both maps are marginalized out by considering Gaussian Process priors. Hence, a CGPDM projects a high-dimensional state space into a smaller dimension latent space, in which it is feasible to learn the system dynamics from training data. The modeling capacity of CGPDM has been tested in both a simulated and a real scenario, where it proved to be capable of generalizing over a wide range of movements and confidently predicting the cloth motions obtained by previously unseen sequences of control actions.

23 March 2026 in IEEE Robotics and Automation Practice
Verified
Journal Article
Closed access
FAFabio Amadio
Fabio Amadio
CDClemente Donoso
Clemente Donoso
DTDionis Totsila
Dionis Totsila
RLRaphael Lorenzo
Raphael Lorenzo
QRQuentin Rouxel
Quentin Rouxel
SISerena Ivaldi
Serena Ivaldi
This paper describes the Inria team's integrated robotics system used in the 1st euROBIN coopetition, during which service robots performed voice-activated household tasks in a kitchen setting. The team developed a modified TIAGo++ platform that leverages a whole-body control stack for autonomous and teleoperated modes, and an LLM-based pipeline for instruction understanding and task planning. The key contributions (opens-sourced) are the integration of these components and the design of custom teleoperation devices, addressing practical challenges in the deployment of service robots.

27 October 2025 in HAL
Verified
Preprint
Open access
PDF file
FAFabio Amadio
Fabio Amadio
HLHongbo Li
Hongbo Li
LULorenzo Uttini
Lorenzo Uttini
DKDimitrios Kanoulas
Dimitrios Kanoulas
SISerena Ivaldi
Serena Ivaldi
EHEnrico Mingo Hoffman
Enrico Mingo Hoffman
Humanoid robots increasingly adopt hybrid serial-parallel kinematics to improve structural stiffness, mass distribution, and impact robustness. However, these mechanisms introduce complexity associated with simulation and control, which affects algorithms for Reinforcement Learning (RL)-based locomotion. This paper presents a case study of an RL end-to-end pipeline that trains walking policies for Kangaroo, a 72 degrees of freedom biped whose legs contain several hybrid serial-parallel chains, without kinematic simplifications. Training is performed using the Isaac Lab framework, leveraging the Isaac Sim built-in constraint capabilities. We perform an ablation study on the observation state to assess the role of redundant information from Kangaroo’s measured state (passive/active joint measurements). Furthermore, a set of trained policies is validated in MuJoCo, demonstrating a degree of robustness to the Sim-to-Sim gap. The closed-loop behaviors of the tested policies successfully transfer in most cases, despite differences in how contacts and constraints are modeled across the two simulators. Finally, we analyzed how minor differences in the action rate penalty weight can deeply affect the locomotion stability of the resulting policies when deployed in a different simulator.

6 September 2022 in arXiv
Verified
Preprint
Open access
PDF file
FAFabio Amadio
Fabio Amadio
ALAlberto Dalla Libera
Alberto Dalla Libera
RARiccardo Antonello
Riccardo Antonello
DNDaniel Nikovski
Daniel Nikovski
RCRuggero Carli
Ruggero Carli
DRDiego Romeres
Diego Romeres
In this paper, we present a Model-Based Reinforcement Learning (MBRL) algorithm named \emph{Monte Carlo Probabilistic Inference for Learning COntrol} (MC-PILCO). The algorithm relies on Gaussian Processes (GPs) to model the system dynamics and on a Monte Carlo approach to estimate the policy gradient. This defines a framework in which we ablate the choice of the following components: (i) the selection of the cost function, (ii) the optimization of policies using dropout, (iii) an improved data efficiency through the use of structured kernels in the GP models. The combination of the aforementioned aspects affects dramatically the performance of MC-PILCO. Numerical comparisons in a simulated cart-pole environment show that MC-PILCO exhibits better data efficiency and control performance w.r.t. state-of-the-art GP-based MBRL algorithms. Finally, we apply MC-PILCO to real systems, considering in particular systems with partially measurable states. We discuss the importance of modeling both the measurement system and the state estimators during policy optimization. The effectiveness of the proposed solutions has been tested in simulation and on two real systems, a Furuta pendulum and a ball-and-plate rig.