Research

Research Interests

My research is in robot learning and physical AI. I develop robots that cope with the complexity of our world: they use cameras, touch, and other sensors to observe and parse their surroundings, and they learn to adjust their behavior accordingly. My group studies imitation learning and reinforcement learning with world models, with a primary focus on robot manipulation. Our aim is to bring autonomous robotic assistance to settings where working conditions are dangerous, strenuous, or impractical for humans, such as manufacturing, construction, agriculture, and assistance to people with limited mobility. My work is at the intersection of robotics and machine learning, which is reflected in my dual appointment within KU Leuven’s departments of electrical and mechanical engineering.

I also have a long-standing soft spot for space robotics, in which I remain active: my group works on visual pose estimation for on-orbit servicing and situational awareness, building on my years at NASA/JPL, where I led machine-vision work for the Mars Sample Return campaign and worked on mobile manipulation for planetary exploration.

My work focuses on casting robotics problems into a form that lends itself to data-driven modeling. Much of the creativity involved here goes into crafting… models that allow us to leverage state-of-the-art machine learning tools to solve a given task. I study multimodal perceptual models that enable artificial decision-making in open-ended environments, with a focus on machine vision and robotics, including planning and control for manipulation, mobility, and task modeling. At the core of these models, I study perception-action embeddings that are capable of expressing the constraints and objectives of a task while maximizing invariance to inconsequential perceptual parameters, with the purpose of generalizing to new data. My scientific interests lie in the development of such representations to address problems that cannot be efficiently modeled through simpler means.

A defining aspect of my work is grounding in sensorimotor experience: I study models that run on physical systems. My Ph.D. and postdoc (ULiege and KTH) focused primarily on manipulation systems composed of industrial arms and grippers equipped with visual and tactile sensors, targeting home or industrial applications. During my tenure at NASA/JPL, I extended my reach to legged/wheeled mobility, situational awareness, and field robots that implement the hardware and software stacks needed to accomplish a real-world task in a relevant open-ended environment. I worked on self-sufficient autonomous mobile manipulation systems and fielded them in extreme natural settings, including caves, deserts, and turbid water. [show more]

Keywords: Robot learning, physical AI, robot manipulation, computer vision, space robotics.

Projects

active

ReHand (2026–2030)

Overcoming hand paralysis with a multistate closed-loop brain-exoskeleton interface.

active

H2XO (2024–2028)

Relieving heavy-duty work with a haptically-controlled robot assistant that learns on the job.

active

NESP (2022–2026)

Neural Estimation of Spacecraft Pose.

active

Neurobotics (2022–2026)

Blended brain-machine control for human assistance using hybrid smart systems.

active

RoBétArmé (2022–2026)

Autonomous robotic concrete projection.

TubePicker (2021–2022)

Detect and retrieve sample tubes on Mars.

SatBuilder (2018–2019)

Autonomous visual manipulation for in-space assembly.

Mars Sample Return (2016–2021)

Return Martian soil samples to Earth.

FreeClimber (2016–2021)

Learning to climb rock walls of unknown geometry.

RCTA/RoMan (2016–2021)

Autonomous manipulation and mobility in unstructured terrain.

InVADER (2016–2021)

In-situ vent analysis divebot for exobiology research.

UMM (2016–2019)

Autonomous mobile manipulation in turbid water.

Plexegen (2012–2016)

Grasping novel objects with tactile-based grasp stability estimates.

Grasp (2011–2015)

Task-oriented Grasping: learning to use new tools and objects.

CogX (2008–2012)

Cognitive Systems that Self-Understand and Self-Extend.

Feature Hierarchies (2006–2010)

Learning of multi-Dimensional, multi-modal features for robotic grasping.

PACO-PLUS (2006–2010)

Perception, action and cognition through learning of object-action complexes.

Ph.D. Students

Current Ph.D. students:

• Alessandra Miuccio (Personal Webpage)
• Yutong Hu (Personal Webpage)
• Ze Fu
• Elias De Smijter
• Yurui Du
• Pinhao Song
• Louis Hanut

Past Ph.D. students:

• Zhongxi Li
• Antoine Legrand (graduated in 2026)
• Deniz Soysal
• Joseph Bowkett (informal co-supervision with Joel Burdick, Caltech)
• Emil Hyttinen (co-supervision with Danica Kragic, KTH)
• Martin Hjelm (informal co-supervision with Danica Kragic, KTH)
• Alexander Rietzler (informal co-supervision with Justus Piater, UIBK)
• Ashley Liddiard (informal co-supervision with Bryan Tripp, UWaterloo)
• Yasemin Bekiroglu (informal co-supervision with Danica Kragic, KTH)

Software

My group’s software is available on our GitLab portal. Older software from my Ph.D. and postdoc is freely available on this page.

Jobs

See this page.

Videos

See this page.

Events

See this page.