Robotics · Perception · Control
Assistant Professor at the University of Zaragoza researching robotics, deformable-object manipulation and shape control.
Field notes
Experience & education
Research agenda
I study how robots perceive, model, and control deformable objects. My work connects shape perception, mapping, and manipulation with a focus on robust, data-efficient methods and multi-robot coordination.
Interactive Demos: For a more accessible introduction to my research, check out the section with interactive visualizations designed for a general audience.
Demos & outreach
Four interactive demonstrations of deformable-object shape control, followed by my educational robotics and outreach work.
Four demonstrations
The demonstrations build up a single control loop: manipulate a deformable object by hand, quantify how close two shapes are, choose the features that define shape error, then estimate the shape Jacobian and close the loop.
The demonstrations are ordered, but each one stands alone.
Step 1 · Manipulate
Two planar manipulators hold a deformable linear object at both ends. Moving either gripper changes the shape of the whole object.
Task: Move one gripper, or its rotation ring, and keep the change only when the error plot goes down.
The same manipulators, rod model and target return in step 4, where an estimated Jacobian replaces the operator.
Explore the research: object-compliant shape control · multiscale 3-D shape control
Step 2 · Compare
Any correction loop needs a scalar that states how far the current shape is from the target. This demonstration makes that measurement explicit.
Task: Remove material, measure again, and undo an edit that lowers the score: the compare-and-correct step a controller repeats.
The score used here is a simple area overlap. Shape control on real deformable objects needs correspondence-aware metrics.
Explore the research: time-consistent surface mapping · elastic contour mapping
Step 3 · Represent
The controller compares numbers, not pictures. Point coordinates, edge vectors and curvature are three ways to turn the same rod into numbers.
Task: Press Translate and watch only the point coordinates move. Press Rotate and watch curvature stay flat. Press Bend and watch all three react.
A feature that ignores a movement is invariant to it. Invariance rejects disturbances you do not care about, and hides directions the controller can no longer correct.
Step 4 · Shape control
The same task without a human in the loop. The robots first explore, then use what they measured as feedback.
Task: Run it once per feature vector and compare: which errors reach zero, and what the robots do differently in each case.
Small test movements give the shape Jacobian, the local relation between gripper motion and feature motion. The loop then inverts it to turn error into motion.
Explore the research: data-driven shape trajectory control
For my PhD thesis presentation, I created a flyer that summarizes the main contributions and findings of my research. Feel free to download and explore:
Selected record