Pietro Falco

Alessandro Adami, Sebastian Zudaire, Ruggero Carli et al.

Constructing physically accurate simulation environments (Real2Sim) traditionally relies on manual system identification or rigid, exhaustive exploration routines. These task-agnostic pipelines often fail to leverage semantic scene context, leading to redundant physical interactions and inefficient data acquisition. In this paper, we present an autonomous, intent-driven Real2Sim framework that leverages Vision-Language Models (VLMs) for Semantic Task Decomposition. Given a high-level natural language request, an incomplete simulation description, and a visual observation, the framework autonomously identifies the minimal subset of missing physical parameters required for the simulation task. It then generates a reactive Behavior Tree (BT) composed of atomic motion and sensing primitives to selectively acquire these parameters through contact-rich robotic interaction. Extensive real-world experiments on a torque-controlled Franka Emika Panda demonstrate that our approach accurately estimates object mass, surface geometry, and derived parameters such as friction. Quantitative evaluations reveal significant operational efficiency gains compared to exhaustive baseline methods, while ablation studies confirm the robustness of the prompt architecture across different state-of-the-art VLMs. Furthermore, the reactive hierarchy of the BT acts as a deterministic safety filter, successfully mitigating generative VLM hallucinations and preventing unsafe physical anomalies. Ultimately, this work provides a scalable, efficient, and interpretable pipeline for building physics-aware digital twins directly from unstructured human intent.

Alessandro Adami, Tommaso Tubaldo, Marco Todescato et al.

Vision-language models (VLMs) have recently demonstrated strong capabilities in mapping multimodal observations to robot behaviors. However, most current approaches rely on end-to-end visuomotor policies that remain opaque and difficult to analyze, limiting their use in safety-critical robotic applications. In contrast, classical robotic systems often rely on structured policy representations that provide interpretability, modularity, and reactive execution. This work investigates how foundation models can be specialized to generate structured robot policies grounded in multimodal perception, bridging high-dimensional learning and symbolic control. We propose a neuro-symbolic approach in which a VLM synthesizes executable Behavior Tree policies from visual observations, natural language instructions, and structured system specifications. To enable scalable supervision without manual annotation, we introduce an automated pipeline that generates a synthetic multimodal dataset of domain-randomized scenes paired with instruction-policy examples produced by a foundation model. Real-world experiments on two robotic manipulators show that structured policies learned entirely from synthetic supervision transfer successfully to physical systems. The results indicate that foundation models can be adapted to produce interpretable and structured robot policies, providing an alternative to opaque end-to-end approaches for multimodal robot decision making.

Niccolò Turcato, Matteo Iovino, Aris Synodinos et al.

Recent advancements in Large Language Models (LLMs) and Visual Language Models (VLMs) have significantly impacted robotics, enabling high-level semantic motion planning applications. Reinforcement Learning (RL), a complementary paradigm, enables agents to autonomously optimize complex behaviors through interaction and reward signals. However, designing effective reward functions for RL remains challenging, especially in real-world tasks where sparse rewards are insufficient and dense rewards require elaborate design. In this work, we propose Autonomous Reinforcement learning for Complex Human-Informed Environments (ARCHIE), an unsupervised pipeline leveraging GPT-4, a pre-trained LLM, to generate reward functions directly from natural language task descriptions. The rewards are used to train RL agents in simulated environments, where we formalize the reward generation process to enhance feasibility. Additionally, GPT-4 automates the coding of task success criteria, creating a fully automated, one-shot procedure for translating human-readable text into deployable robot skills. Our approach is validated through extensive simulated experiments on single-arm and bi-manual manipulation tasks using an ABB YuMi collaborative robot, highlighting its practicality and effectiveness. Tasks are demonstrated on the real robot setup.

Cristina Iacono, Mariarosaria Meola, Federica Conte et al.

Cholecystectomy is one of the most frequently performed procedures in gastrointestinal surgery, and the laparoscopic approach is the gold standard for symptomatic cholecystolithiasis and acute cholecystitis. In addition to the advantages of a significantly faster recovery and better cosmetic results, the laparoscopic approach bears a higher risk of bile duct injury, which has a significant impact on quality of life and survival. To avoid bile duct injury, it is essential to improve the intraoperative visualization of the bile duct. This work aims to address this problem by leveraging a deep-learning approach for the localization of the biliary tract from white-light images acquired during the surgical procedures. To this end, the construction and annotation of an image database to train the Yolo detection algorithm has been employed. Besides classical data augmentation techniques, the paper proposes Generative Adversarial Network (GAN) for the generation of a synthetic portion of the training dataset. Experimental results have been discussed along with ethical considerations.

Shahbaz Abdul Khader, Hang Yin, Pietro Falco et al.

Deep reinforcement learning (DRL) has been successfully used to solve various robotic manipulation tasks. However, most of the existing works do not address the issue of control stability. This is in sharp contrast to the control theory community where the well-established norm is to prove stability whenever a control law is synthesized. What makes traditional stability analysis difficult for DRL are the uninterpretable nature of the neural network policies and unknown system dynamics. In this work, stability is obtained by deriving an interpretable deep policy structure based on the $\textit{energy shaping}$ control of Lagrangian systems. Then, stability during physical interaction with an unknown environment is established based on $\textit{passivity}$. The result is a stability guaranteeing DRL in a model-free framework that is general enough for contact-rich manipulation tasks. With an experiment on a peg-in-hole task, we demonstrate, to the best of our knowledge, the first DRL with stability guarantee on a real robotic manipulator.

Matteo Iovino, Jonathan Styrud, Pietro Falco et al.

Modern industrial applications require robots to be able to operate in unpredictable environments, and programs to be created with a minimal effort, as there may be frequent changes to the task. In this paper, we show that genetic programming can be effectively used to learn the structure of a behavior tree (BT) to solve a robotic task in an unpredictable environment. Moreover, we propose to use a simple simulator for the learning and demonstrate that the learned BTs can solve the same task in a realistic simulator, reaching convergence without the need for task specific heuristics. The learned solution is tolerant to faults, making our method appealing for real robotic applications.

Shahbaz Abdul Khader, Hang Yin, Pietro Falco et al.

Reinforcement Learning (RL) of robotic manipulation skills, despite its impressive successes, stands to benefit from incorporating domain knowledge from control theory. One of the most important properties that is of interest is control stability. Ideally, one would like to achieve stability guarantees while staying within the framework of state-of-the-art deep RL algorithms. Such a solution does not exist in general, especially one that scales to complex manipulation tasks. We contribute towards closing this gap by introducing $\textit{normalizing-flow}$ control structure, that can be deployed in any latest deep RL algorithms. While stable exploration is not guaranteed, our method is designed to ultimately produce deterministic controllers with provable stability. In addition to demonstrating our method on challenging contact-rich manipulation tasks, we also show that it is possible to achieve considerable exploration efficiency--reduced state space coverage and actuation efforts--without losing learning efficiency.

Shahbaz A. Khader, Hang Yin, Pietro Falco et al.

Reinforcement learning (RL) has had its fair share of success in contact-rich manipulation tasks but it still lags behind in benefiting from advances in robot control theory such as impedance control and stability guarantees. Recently, the concept of variable impedance control (VIC) was adopted into RL with encouraging results. However, the more important issue of stability remains unaddressed. To clarify the challenge in stable RL, we introduce the term all-the-time-stability that unambiguously means that every possible rollout will be stability certified. Our contribution is a model-free RL method that not only adopts VIC but also achieves all-the-time-stability. Building on a recently proposed stable VIC controller as the policy parameterization, we introduce a novel policy search algorithm that is inspired by Cross-Entropy Method and inherently guarantees stability. Our experimental studies confirm the feasibility and usefulness of stability guarantee and also features, to the best of our knowledge, the first successful application of RL with all-the-time-stability on the benchmark problem of peg-in-hole.

Pietro Falco, Shuang Lu, Ciro Natale et al.

In this work, we introduce the problem of cross-modal visuo-tactile object recognition with robotic active exploration. With this term, we mean that the robot observes a set of objects with visual perception and, later on, it is able to recognize such objects only with tactile exploration, without having touched any object before. Using a machine learning terminology, in our application we have a visual training set and a tactile test set, or vice versa. To tackle this problem, we propose an approach constituted by four steps: finding a visuo-tactile common representation, defining a suitable set of features, transferring the features across the domains, and classifying the objects. We show the results of our approach using a set of 15 objects, collecting 40 visual examples and five tactile examples for each object. The proposed approach achieves an accuracy of 94.7%, which is comparable with the accuracy of the monomodal case, i.e., when using visual data both as training set and test set. Moreover, it performs well compared to the human ability, which we have roughly estimated carrying out an experiment with ten participants.

Pietro Falco, Matteo Saveriano, Eka Gibran Hasany et al.

In this paper, we present a descriptor for human whole-body actions based on motion coordination. We exploit the principle, well known in neuromechanics, that humans move their joints in a coordinated fashion. Our coordination-based descriptor (CODE) is computed by two main steps. The first step is to identify the most informative joints which characterize the motion. The second step enriches the descriptor considering minimum and maximum joint velocities and the correlations between the most informative joints. In order to compute the distances between action descriptors, we propose a novel correlation-based similarity measure. The performance of CODE is tested on two public datasets, namely HDM05 and Berkeley MHAD, and compared with state-of-the-art approaches, showing recognition results.

Pietro Falco, Abdallah Attawia, Matteo Saveriano et al.

In this letter, we present an approach for learning in-hand manipulation skills with a low-cost, underactuated prosthetic hand in the presence of irreversible events. Our approach combines reinforcement learning based on visual perception with low-level reactive control based on tactile perception, which aims to avoid slipping. The objective of the reinforcement learning level consists not only in fulfilling the in-hand manipulation goal, but also in minimizing the intervention of the tactile reactive control. This way, the occurrence of object slipping during the learning procedure, which we consider an irreversible event, is significantly reduced. When an irreversible event occurs, the learning process is considered failed. We show the performance in two tasks, which consist in reorienting a cup and a bottle only using the fingers. The experimental results show that the proposed architecture allows reaching the goal in the Cartesian space and reduces significantly the occurrence of object slipping during the learning procedure. Moreover, without the proposed synergy between reactive control and reinforcement learning it was not possible to avoid irreversible events and, therefore, to learn the task.

Shahbaz Abdul Khader, Hang Yin, Pietro Falco et al.

In this letter, we investigate learning forward dynamics models and multi-step prediction of state variables (long-term prediction) for contact-rich manipulation. The problems are formulated in the context of model-based reinforcement learning (MBRL). We focus on two aspects-discontinuous dynamics and data-efficiency-both of which are important in the identified scope and pose significant challenges to State-of-the-Art methods. We contribute to closing this gap by proposing a method that explicitly adopts a specific hybrid structure for the model while leveraging the uncertainty representation and data-efficiency of Gaussian process. Our experiments on an illustrative moving block task and a 7-DOF robot demonstrate a clear advantage when compared to popular baselines in low data regimes.