Carolina Higuera

I am a 4th-year Ph.D. student in the Paul G. Allen School at the University of Washington, where I am advised by Byron Boots. Additionally, I am a visiting researcher at Meta, working with Mustafa Mukadam as part of the AIM mentorship program.

I received my M.S. in Electronics Engineering from Universidad de los Andes, Bogota, Colombia. There, I worked on multiagent reinforcement learning for traffic light signal control.

I am honored to be a Fulbright scholar from Colombia 🇨🇴

Contact:

• chiguera [at] cs [dot] washington [dot] edu
• carohiguera [at] meta [dot] com

Email  /  Google Scholar

My research focuses on developing models that enable robot manipulators to interpret their environment through vision-based tactile perception. Specifically, I've been exploring how tracking extrinsic contacts between objects and their environment can be advantageous for policy learning. However, to truly leverage touch in robotics, we need a common backbone. Currently, I'm working on learning tactile representations using self-supervised learning that can be versatile for both static and dynamic contact interactions via vision-based tactile sensing

• Visitor Researcher at FAIR and GUM team, advised by Mustafa Mukadam.
• Summer 2023 research internship at FAIR with the GUM team and advised by Mustafa Mukadam.
• NCF paper accepted at ICRA 2023

Sparsh is a family of general touch representations trained via self-supervision algorithms such as MAE, DINO and JEPA. Sparsh is able to generate useful representations for DIGIT, Gelsight'17 and Gelsight Mini. It outperforms end-to-end models in the downstream tasks proposed in TacBench by a large margin, and can enable data efficient training for new downstream tasks.

Improve NCF to enable sim-to-real transfoer and use it to train policies for insertion tasks. We demonstrate the utility of extrinsisc contacts during policy learning and perform experiments on a real tasks.

Best Paper Award, ICRA 2023 Workshop on Effective Representations, Abstractions, and Priors for Robot Learning, (RAP4Robots)

We propose Tactile Diffusion to bridge the sim-to-real gap when using vision-based tactile sensors, like DIGIT. We demonstrate the utility of Tactile Diffusion on zero-shot clasification of Braille characters.

Neural Contact Fields are an implicit representation for tracking extrinsic contact on an object surface (between object and environment) with vision-based tactile sensing (between robot hand and object).

Application of Deep RL for an Energy Management System (EMS) and its comparison with respect to classical techniques such as Rule-Based and Model Predictive Control.

Application of MARL to traffic light signal control to reduce travel time. We simulated a network with six signalized intersections in SUMO, using real data from the Transit Department of Bogota, Colombia. This project was my Master thesis!

A system to analyze urban traffic using computer vision to get realiable information of traffic flow in Bogota, Colombia.

template