Research interests

Doctoral focus

Self-Sustaining Machines and Long Term Autonomy

My doctoral work at IE University’s Cyber-Physical Life Lab asks how a fleet of robots can keep itself running where human technicians are scarce, slow to arrive or at risk, such as large solar farms and nuclear decommissioning sites. Robots can already monitor their own condition, but a person still diagnoses the fault, carries the spare part and makes the swap, and I work on closing that loop. Each robot detects its own wear from the gap between what it commands and what its actuators actually do. I use disturbance-observer residuals for this, because they should separate wear from whatever task the robot is doing. Its parts are designed so that another robot can replace them, the way a car is built for a mechanic. I compare mechanical latches with joints that lock and release on command when a magnetic field or heat is applied. After a swap, the robot loads the software for the new part and confirms that its controllers are still stable before it goes back to work. At the fleet level, I study how a few servicers, which wear out too, should decide which robot to service and when. I also study when robotic repair beats human repair on cost, time and safety. Parts that come out are tested and reused where possible, so the fleet keeps both its robots and its parts in service. The work starts with a testbed of the lab’s ROSbots under accelerated wear. It builds toward MARC, a colony of robots assembled from a small set of standard parts that can repair itself and change roles.

Self-sustaining robots Self-repair Modular mechatronic design Preventive maintenance Long-term autonomy
Foundation

Internal-Model & Disturbance-Observer Design

Robust control for uncertain dynamic systems. I designed a canonical internal-model filter that embeds known disturbance dynamics into a DOB-style output-regulation controller asymptotically cancelling matched disturbances while stabilizing an uncertain integrator-chain plant under bounded uncertainty, with stability guarantees proven rather than tuned. This analysis background is what I bring to validating the self-sustaining systems in my doctoral work.

Internal-model principle Disturbance observers Output regulation Robust stability
Applied interest

Vision-Language-Action Models

VLA models for robotic manipulation, made sample-efficient and robust enough for modest hardware and real robots. A recent field study injected point-cloud data into the π0 VLA family on an SO-101 arm, cutting frame-edge failure from 40% to 10% under appearance and viewpoint shift using a single 12 GB GPU, 70 real-robot demonstrations, and ~0.15% of parameters trainable.

Vision-language-action Manipulation Point-cloud injection Sample efficiency