Institution profile

Istituto Italiano di Tecnologia

Academic institutioneurope · it
Official website
Research library337linked papers
Opportunities0open roles
Selected work

Representative Papers

Uncoordinated access to serverless computing in MEC systems for IoT

May 01, 2020Comput. Networks

To address the low-latency access challenge for lightweight IoT tasks in multi-access edge computing (MEC), this paper proposes a coordination-free edge serverless computing access mechanism that eliminates centralized scheduling and enables autonomous, asynchronous, low-latency invocation of edge functions by end devices. Methodologically, it introduces three core innovations: (1) a probabilistic distributed resource reservation strategy, (2) a lightweight function registration/discovery protocol, and (3) a decentralized invocation mechanism—implemented atop OpenFaaS and a Kubernetes-based edge extension framework. Evaluation on a real-world MEC-IoT testbed demonstrates that the proposed mechanism reduces average access latency by 47%, increases throughput by 3.2×, and cuts control-plane signaling overhead by 89%, effectively alleviating scheduling bottlenecks induced by network and workload heterogeneity.

14 citations1 influentialRead paper

Exo-Muscle: A Semi-Rigid Assistive Device for the Knee

Oct 01, 2021IEEE Robotics and Automation Letters

Addressing two critical bottlenecks—misalignment of the knee joint’s anatomical rotation center in rigid exoskeletons and uncertain force transmission in soft tendon-driven systems—this study proposes a semi-rigid extramuscular walking aid. The design innovatively employs a semi-rigid chain mechanism that synergistically integrates the kinematic precision of rigid structures with the wearability and adaptability of soft systems. Anatomical-axis alignment is achieved via human kinematic modeling, eliminating rotational center offset entirely. By integrating tendon actuation with mechatronic co-design, the system ensures deterministic and repeatable load compensation. Experimental results demonstrate stable assistive torque output up to 38 Nm across the full knee flexion–extension range, enabling precise, controllable assistance. This significantly enhances human–robot cooperation performance and user comfort.

7 citations1 influentialRead paper

Information-Theoretic Detection of Bimanual Interactions for Dual-Arm Robot Plan Generation

May 01, 2025IEEE Robotics and Automation Letters

This work addresses the challenge of efficiently generating executable plans for dual-arm robotic tasks from human demonstrations, where bimanual coordination strategies are often complex and difficult to model. The authors propose a novel approach that leverages a single RGB video demonstration to synthesize structured, modular behavior tree plans. Their method uniquely integrates Shannon information theory to analyze information flow between hands, scene graph parsing to extract action semantics, and one-shot learning to enable generalization. By unifying these components within a behavior tree framework, the approach produces adaptable execution plans without requiring extensive training data. Evaluated on both a newly curated dataset and existing public benchmarks, the method demonstrates significant performance gains over current state-of-the-art techniques, marking a notable advance in centralized bimanual coordination planning.

4 citationsRead paper

Ice-Breakers, Turn-Takers and Fun-Makers: Exploring Robots for Groups with Teenagers

Aug 29, 2022IEEE International Symposium on Robot and Human Interactive Communication

This study investigates how social robots can support adolescent group interactions to foster identity development and self-esteem. We conducted a two-week summer camp employing participatory methods—including focus groups, in-depth interviews, adolescent-led co-design sessions (10+ hours), and Wizard-of-Oz prototype testing—to systematically uncover dynamic interaction needs across ice-breaking, turn-taking, and engagement-fostering scenarios. To our knowledge, this is the first long-term, adolescent-centered co-design study of social robots for group settings. Findings reveal adolescents’ expectations of robot roles form a dynamic spectrum, necessitating adaptive functionality aligned with group developmental stages (forming → norming → performing). We identify three context-dependent core assistive functions, empirically demonstrate adolescents’ capacity to actively reinterpret and reconfigure robot roles, and propose a transferable “group–robot interaction stage model” alongside four evidence-based design principles.

4 citationsRead paper

Real-time Fall Prevention system for the Next-generation of Workers

May 30, 2025

Early detection of fall risk among healthy, physically robust workers in industrial settings remains challenging due to the scarcity of real-world fall incidents for model training. Method: This paper proposes a novel wearable real-time fall prevention paradigm integrating physics-informed modeling with deep learning. We formulate an inverted-pendulum-based dynamical model to abstract the fall process and generate large-scale, diverse synthetic training data—circumventing the paucity of authentic fall samples. A lightweight temporal neural network is designed for low-latency signal processing and risk classification on embedded edge devices. Contribution/Results: We introduce the first “physics-guided, data-driven” co-design framework, markedly enhancing model generalizability and robustness. Experimental evaluation under simulated industrial dynamic conditions demonstrates a false alarm rate below 3% and end-to-end response latency under 200 ms, establishing a scalable technical foundation for generic wearable fall-prevention systems.

3 citationsRead paper
Recent publications

Latest Papers

How to Learn from What a Human Would Avoid? Intervention-Aware World Models with Real-World RL for Dexterous Manipulation

Sep 05, 2026

Multi-fingered dexterous manipulation remains a frontier for real-world reinforcement learning (RL) due to the high-dimensional action space and the prohibitive cost of hardware failures. While human-in-the-loop (HIL) RL allows operators to intervene before failures occur, current pipelines often treat these interventions as reactive corrections, discarding the rich safety signal inherent in the operator's decision to take control. In this paper, we ask: How can we learn from what a human would avoid? We present WHIRL, a safety-aware RL framework that transforms binary human interventions into forward-predictive signals for proactive risk avoidance. Our approach centers on an intervention-aware latent world model with four prediction heads: dynamics, reward, termination, and a novel per-state intervention-probability head that learns to predict the likelihood of a human takeover at future states. This head provides an actor-side risk-shaping term that discourages the policy from entering"intervention-prone"regions, modeling the operator's internal safety threshold. We evaluate our framework on a 16-DoF LEAP Hand across tasks spanning convex and irregular object grasping, prismatic manipulation, and long-horizon multi-stage tasks. Our results show that predictive risk-shaping enables the system to achieve a 96.7 percent success rate on complex grasping tasks while reducing the operator intervention burden by up to 84 percent in step-weighted terms. By closing the loop between human intuition and predictive world modeling, this work provides a practical safety-aware recipe for training complex dexterous agents in the real world while reducing operator fatigue and hardware-risk exposure.

0 citationsRead paper