Human-Inspired robot learning

January 3, 2026

Overview

Humans are remarkably skilled at manipulating objects, adapting their movements to the task, the context and the objects at hand with little apparent effort. Our research starts from the conviction that, for robots to operate alongside people and act in a way that feels natural and predictable, they should be designed and trained by taking inspiration from the principles that govern human motor control. We study how humans plan, control and coordinate their movements, and translate these insights into planning, control and learning guidelines for robots. This spans the analysis of human manipulation and motor synergies, the design of human-like motion and grasping strategies, and the development of data-efficient learning methods that let robots acquire manipulation skills from a handful of demonstrations rather than from massive task-specific datasets.

Key results

IEEE IROS26

GAP: Geometric Anchor Pre-training for Data-Efficient Visuomotor Learning of Manipulation TasksDistils visual observations into stable geometric anchors, so a frozen vision foundation model learns new manipulation tasks from about a dozen demonstrations.

Paper

Under review

GORDON: Graph-based Object-centric Rewards for Decomposition of Long-Horizon ManipulationLearns dense rewards for long-horizon manipulation from action-free videos, using object-centric graphs to track task progress and discover subtasks automatically.

Paper

IEEE/RSJ IROS23

Domain Randomization for Robust, Affordable and Effective Closed-loop Control of Soft RobotsDomain randomization makes closed-loop control of soft robots robust and effective, narrowing the gap between simulation and the real world.

Paper

IEEE RA-L

A deep neural architecture for few-shot learning of novel human-inspired grasp strategies from RGB dataLearns new human-inspired grasp strategies from only a few examples.

Paper

IEEE ICRA20

A technical framework for human-like motion generation with autonomous anthropomorphic redundant manipulatorsA framework that makes anthropomorphic robot arms move in a human-like way, so their motions look natural and predictable to people.

Paper

IEEE RA-L

Learning from humans how to grasp: a data-driven architecture for autonomous grasping with an anthropomorphic soft handA data-driven architecture that learns from human demonstrations how to grasp objects autonomously with an anthropomorphic soft hand.

Paper

Key Findings

A recurring theme in our work is that compact, structured representations inspired by human motor behaviour make robot learning far more efficient and robust. By distilling visual observations into stable geometric anchors, we show that a frozen vision foundation model can be adapted to new manipulation tasks from as few as a dozen demonstrations, while remaining resilient to scene perturbations and domain shift. In parallel, we demonstrated that off-the-shelf vision-language models can support high-level robot planning without any task-specific training, drastically reducing the amount of data needed to deploy a robot in a new environment. Together with our earlier results on synergy-based grasping and human-like motion planning, these findings point towards robots that learn the way humans do — reusing prior knowledge and a small number of examples instead of brute-force data collection.

Robi: bringing human-inspired manipulation to the real world

These ideas are being put to the test in Robi, an industrial collaboration in which we develop a humanoid robot able to perform complex, dexterous manipulation tasks — such as preparing and serving cocktails — in a real operational setting. Robi is a great playground for our human-inspired approach: it requires fluid, human-like coordination, robust perception, and the ability to learn and adapt skills efficiently, all while operating safely in an environment shared with people.

Challenges and Learnings

Working on human-inspired manipulation, and on a real platform like Robi, has taught us that the gap between simulation and the real world remains one of the hardest challenges: contacts, friction and the variability of real objects are difficult to model, and policies that work in the lab can be brittle once deployed. We have learned that data efficiency and robustness, rather than raw accuracy on a single benchmark, are the properties that matter most for robots acting in the wild. Designing inductive biases inspired by human motor control — structured representations, reusable skills, and human-like coordination — consistently helps us obtain behaviours that are more sample-efficient, more interpretable, and more reliable when the robot leaves the controlled environment.

Papers

Book

Human-aware robotics: Modeling human motor skills for the design, planning, and control of a new generation of robotic devices

Monograph — STAR series, SpringerAverta, G.

Robot Planning & Control

GAP: Geometric Anchor Pre-training for Data-Efficient Visuomotor Learning of Manipulation Tasks

IROS

IEEE IROS26Buoso, D., Protopapa, A., Di Carlo, S., Pistilli, F., Averta, G.

GORDON: Graph-based Object-centric Rewards for Decomposition of Long-Horizon Manipulation

(Under Review)Protopapa, A., Buoso, D., Pistilli, F., Chalvatzaki, G., Averta, G.

On the estimation of object inertial parameters through optimal manipulation patterns

T-SMC

IEEE T-SMCMarcello, L., Napolitano, O., Bonifati, P., Averta, G., Bianchi, M., Salaris, P.

Nail It! a learNing framework for Autonomous surgIcaL suturIng and Teleoperation on the dVRK

IJCARSMuratore, L., Pescio, M., Barontini, F., Marzola, F., Leoncini, P., Ammirati, C.A., Arezzo, A., Dagnino, G., Averta, G.

A TaskGraspNet: A Large Scale Benchmark For Task-Oriented Grasping

(Under Review)Robbiano, L., Falco, G.S., Pistilli, F., Averta, G.

RF-DROPO: Data-Efficient Adaptive Domain Randomization for Zero-Shot Sim-to-Real Transfer of Soft Robot Control Policies

(Under Review)Protopapa, A., Tiboni, G., Vandaele, J., Tommasi, T., Camoriano, R., Duriez, C., Averta, G.

Select2Plan: Training-Free ICL-Based Planning through VQA and Memory Retrieval

RA-L

IEEE RA-LBuoso, D., Robinson, L., Averta, G., Torr, P., Franzmeyer, T., De Martini, D.

A general approach for generating artificial human-like motions from functional components of human upper limb movements

Control Eng. Pract.

Elsevier Control Eng. PracticeBaracca, M., Averta, G., Bianchi, M.

Domain Randomization for Robust, Affordable and Effective Closed-loop Control of Soft Robots

IROS

IEEE/RSJ IROS23Tiboni, G., Protopapa, A., Tommasi, T., Averta, G.

Online vs. Offline Adaptive Domain Randomization Benchmark

HFR22Tiboni, G., Arndt, K., Averta, G., Kyrki, V., Tommasi, T.

A deep neural architecture for few-shot learning of novel human-inspired grasp strategies from RGB data

RA-L

IEEE RA-LCollodi, L., Bacciu, D., Bianchi, M., Averta, G.

Understanding human manipulation with the environment: a novel taxonomy for video labelling

RA-L

IEEE RA-LArapi, V., Della Santina, C., Averta, G., Bicchi, A., Bianchi, M.

Toward human-like motion generation in the Cartesian domain with robotic manipulators

I-RIM 3D21Baracca, M., Bonifati, P., Nisticò, Y., Catrambone, V., Valenza, G., Bicchi, A., Averta, G., Bianchi, M.

Learning to prevent grasp failure with soft hands: from on-line prediction to dual-arm grasp recovery

Adv. Intell. Syst.

Wiley Advanced Intelligent SystemsAverta, G., Barontini, F., Valdambrini, I., Cheli, P., Bacciu, D., Bianchi, M.

On the Planning, Control, and Perception of Soft Robotic End-Effectors

Frontiers in Robotics and AIAverta, G., Della Santina, C., Ficuciello, F., Roa, M.A., Bianchi, M.

Enhancing Robot Physical Interaction via Optimal Impedance Profiles

IEEE RAS Biorob20Averta, G., Hogan, N.

Modeling Previous Trial Effect in Human Manipulation Through Iterative Learning Control

Adv. Intell. Syst.

Advanced Intelligent SystemsCenceschi, L.*, Della Santina, C.*, Averta, G., Garabini, M., Fu, Q., Santello, M., Bianchi, M., Bicchi, A.

Deep learning techniques for modelling human manipulation and its translation for autonomous robotic grasping with soft end-effectors

"AI for Emerging Verticals: Human-Robot Computing, Sensing and Networking" — IETArapi, V., Zhang, Y., Averta, G., Della Santina, C., Bianchi, M.

To grasp or not to grasp: an end-to-end deep-learning approach for predicting grasping failures in soft hands

IEEE RoboSoft20Arapi, V., Zhang, Y., Averta, G., Catalano, M.G., Rus, D., Della Santina, C., Bianchi, M.

A technical framework for human-like motion generation with autonomous anthropomorphic redundant manipulators

ICRA

IEEE ICRA20Averta, G., Caporale, D., Della Santina, C., Bicchi, A., Bianchi, M.

Modeling Human Motor Skills to Enhance Robots' Physical Interaction

HFR20Averta, G., Arapi, V., Bicchi, A., Della Santina, C., Bianchi, M.

Exploiting upper-limb functional synergies for human-like motion generation of anthropomorphic robots

JNER

Journal of NeuroEngineering and RehabilitationAverta, G., Della Santina, C., Valenza, G., Bianchi, M., Bicchi, A.

Human-inspired motion primitives for a safe and effective HRI and autonomous manipulation

I-RIM 3D20Averta, G., Bianchi, M.

Embedding human skills in humanoid manipulators

I-RIM 3D19Averta, G., Bicchi, A., Bianchi, M.

Intelligent Machines for Autonomous Robotics and Manipulation: Combining Soft Robotics, Deep Learning and Human Inspiration

I-RIM 3D19Della Santina, C., Averta, G., Arapi, V., Settimi, A., Bacciu, D., Catalano, M.G., Bicchi, A., Bianchi, M.

Learning from humans how to grasp: a data-driven architecture for autonomous grasping with an anthropomorphic soft hand

RA-L

IEEE RA-LDella Santina, C., Arapi, V., Averta, G., Damiani, F., Fiore, G., Settimi, A., Catalano, M.G., Bacciu, D., Bicchi, A., Bianchi, M.

On the Role of Postural Synergies for Grasp Force Generation and Upper Limb Motion Control

ICNR18Averta, G., Angelini, F., Bicchi, A., Valenza, G., Bianchi, M.

A synergistic behavior underpins human hand grasping force control during environmental constraint exploitation

ICNR18Averta, G., Battaglia, E., Della Santina, C., Catalano, M.G., Bianchi, M.

Incrementality and Hierarchies in the Enrollment of Multiple Synergies for Grasp Planning

RA-L

IEEE RA-LAverta, G., Angelini, F., Bonilla, M., Bianchi, M., Bicchi, A.

Efficient Walking Gait Generation via Principal Component Representation of Optimal Trajectories: Application to a Planar Biped Robot with Elastic Joints

RA-L

IEEE RA-LGasparri, G.M., Manara, S., Caporale, D., Averta, G., Bonilla, M., Marino, H., et al., Garabini, M.

From Humans to Robots: the Role of Cutaneous Impairment in Human Environmental Constraint Exploitation to Inform the Design of Robotic Hands

Award

ICUMT17 — Best Student Paper Award, Best Paper in Session RoboticsAverta, G., Della Santina, C., Battaglia, E., Ciotti, S., et al., Bianchi, M.

Touch-Based Grasp Primitives for Soft Hands: Applications to Human-to-Robot Handover Tasks and Beyond

ICRA

IEEE ICRA17Bianchi, M., Averta, G., Battaglia, E., Rosales, C., et al., Bicchi, A.

Design of an under-actuated wrist based on adaptive synergies

ICRA

IEEE ICRA17Casini, S., Tincani, V., Averta, G., Poggiani, M., Della Santina, C., Battaglia, E., et al., Bicchi, A.

Human Motion Control

A multi-modal under-sensorized wearable system for optimal kinematic and muscular tracking of human upper limb motion

SensorsBonifati, P., Baracca, M., Menolotto, M., Averta, G., Bianchi, M.

Optimal Reconstruction of Human Motion From Scarce Multimodal Data

T-HMS

IEEE Trans. Human-Machine SystemsAverta, G., Iuculano, M., Salaris, P., Bianchi, M.

Estimation of whole-body muscular activation from an optimal set of scarce electromyographic recordings

HFR22Baracca, M., Averta, G., Bianchi, M.

A low-dimensional representation of arm movements and hand grip forces in post-stroke individuals

Scientific Reports

Scientific ReportsKanzler, C.M.*, Averta, G.*, Schwarz, A., Held, J.P.O., Gassert, R., Bicchi, A., Santello, M., Lambercy, O., Bianchi, M.

Optimal electromyographic sensing for whole-body muscular activity estimation

I-RIM 3D22Baracca, M., Averta, G., Bianchi, M.

μ-Limb: A multi-modal, multi-center database on human upper limb motion control in physiological and post-stroke conditions

GigaScience

GigaScienceAverta, G., et al.

Toward brain–heart computer interfaces: a study on the classification of upper limb movements using multisystem directional estimates

J. Neural Eng.

Journal of Neural EngineeringCatrambone, V., Averta, G., Bianchi, M., Valenza, G.

Functional analysis of upper-limb movements in the Cartesian domain

ICNR20Baracca, M., Bonifati, P., Nisticò, Y., Catrambone, V., Valenza, G., Bicchi, A., Averta, G., Bianchi, M.

A novel mechatronic system for evaluating elbow muscular spasticity relying on Tonic Stretch Reflex Threshold estimation

IEEE EMBC20Averta, G., Abbinante, M., Orsini, P., Felici, F., Lippi, P., Bicchi, A., Catalano, M.G., Bianchi, M.

Kineto-dynamic modeling of human upper limb for robotic manipulators and assistive applications

"Advances in Motor Neuroprostheses" — Springer SeriesAverta, G., Bettelani, G.C., Della Santina, C., Bianchi, M.

A functional analysis-based approach to quantify upper limb impairment level in chronic stroke patients: A pilot study

IEEE EMBC19Schwarz, A., Averta, G., Veerbeek, J.M., Luft, A., Held, J.P., Valenza, G., Bicchi, A., Bianchi, M.

On the time-invariance properties of upper limb synergies

TNSRE

IEEE Trans. Neural Systems and Rehabilitation EngineeringAverta, G., Valenza, G., Catrambone, V., Barontini, F., Scilingo, E.P., Bicchi, A., Bianchi, M.

EEG Processing to Discriminate Transitive-Intransitive Motor Imagery Tasks: Preliminary Evidences using Support Vector Machines

IEEE EMBC18Catrambone, V., Greco, A., Averta, G., Bianchi, M., Vanello, N., Bicchi, A., Valenza, G., Scilingo, E.P.

EEG Complexity Maps to Characterize Brain Dynamics during Upper Limb Motor Imagery

IEEE EMBC18Catrambone, V., Greco, A., Averta, G., Bianchi, M., Bicchi, A., Scilingo, E.P., Valenza, G.

Predicting object-mediated gestures from brain activity: an EEG study on gender differences

TNSRE

IEEE Trans. Neural Systems and Rehabilitation EngineeringCatrambone, V., Greco, A., Averta, G., Bianchi, M., Valenza, G., Scilingo, E.P.

Postural Hand Synergies during Environmental Constraint Exploitation

Frontiers in neuroroboticsDella Santina, C., Bianchi, M., Averta, G., Ciotti, S., et al., Bicchi, A.

Unvealing the Principal Modes of Human Upper Limb Movements through Functional Analysis

Frontiers in Robotics and AIAverta, G., Della Santina, C., Battaglia, E., Felici, F., Bianchi, M., Bicchi, A.