Hi, I'm Mario.
I'm a doctoral fellow at the Center for Learning Systems, working with Marc Pollefeys in the Computer Vision and Geometry Group at ETH Zürich and with Wieland Brendel's lab at the ELLIS Institute in Tübingen.
I focus on methods that let autonomous agents learn actionable world representations, flexible enough for complex, goal-oriented behaviour and for autonomously improving their capabilities. In particular, I'm interested in loco-manipulation, to make robots useful both as companions in daily life and in dangerous, high-stakes situations.
Before that I did my master's in Robotics, Systems and Control at ETH, focused on planning and AI, with my thesis at the Prorok Lab in Cambridge. Along the way I worked at the Robotic Systems Lab (trajectory optimization on Gaussian Splats for ANYmal, a rope gripper), at Gravis Robotics (planning and MPC for autonomous excavators), and at TU Darmstadt during my bachelor's in computer science (robot juggling, tactile sensing).
papers
pressing objects to tell how hard they are
Investigating Active Sampling for Hardness Classification with Vision-Based Tactile Sensors
A robot with a tactile sensor tells objects apart by hardness at 88.78% accuracy, compared to 48% for people.