Antonia Bronars

I am a PhD student at MIT working with Professor Pulkit Agrawal in the Improbable AI lab. I am currently investigating how tactile perception can enable more accurate and dexterous robotic manipulation. I did my Masters with Professor Alberto Rodriguez in the MCube Lab, and have also spent time at Mitsubishi Electric Research Labs (MERL).

Prior to MIT, I got my bachelor's in Mechanical Engineering at UC Berkeley. I won the Steidel Award for undergraduate research, working with Professor Oliver O'Reilly on nonlinear dynamics, and Professor Alice Agogino on tensegrity robots.

Email  /  CV  /  LinkedIn  /  Twitter  /  Google Scholar

profile photo
Selected Projects

I am interested in physical robot intelligence, and in particular, the role of tactile feedback in reactive control of dexterous manipulation. My recent work has centered on perception algorithms that estimate object pose from tactile images, to enable precise manipulation tasks.

TEXterity: Tactile Extrinsic deXterity
Antonia Bronars*, Sangwoon Kim*, Alberto Rodriguez
International Conference on Robotics and Automation (ICRA) , 2024
video / arXiv

Our framework estimates and tracks object pose using tactile images while simultaneously generating motion plans to achieve manipulation objectives.

High-Accuracy Tactile Pose Estimation for Small Parts Assembly (coming soon!)
Antonia Bronars, Devesh Jha, Radu Corcodel
In preparation

We assemble tight-tolerance industrial connectors using a only simple controller, without the need for specialized search or insertion algorithms. This is possible via sub-millimeter accuracy tactile pose estimation.

simPLE: a visuotactile method learned in simulation to precisely pick, localize, regrasp, and place objects
Maria Bauza, Antonia Bronars, Yifan Hou, Ian Taylor, Nikhil Chavan-Dafle, Alberto Rodriguez
Science Robotics , 2023
video / arXiv

simPLE learns to pick, regrasp, and place objects precisely given the object CAD model. We develop three main components: task-aware grasping, visuotactile perception, and regrasp planning.

Tac2Pose: Tactile Object Pose Estimation from the First Touch
Antonia Bronars*, Maria Bauza*, Alberto Rodriguez
International Journal of Robotics Research (IJRR), 2022
arXiv

Given an object's CAD model, we learn a tailored perception model in simulation that estimates a probability distribution over possible object poses given a real tactile observation. Our solution is used by Magna, ABB, and MERL.

Gliding Motions of a Rigid Body: The Curious Dynamics of Littlewood's Rolling Hoop
Antonia Bronars, Oliver O'Reilly
Proceedings of the Royal Society A, 2019
paper

We analyze the self-induced jumping behavior of a rolling hoop, and show that the paradoxical motions - first noted by mathematician John Littlewood - are consistent with the principles of mechanics.

Inclined Surface Locomotion Strategies for Spherical Tensegrity Robots
Lee-Huang Chen, Brian Cera, Edward Zhu, Riley Edmunds, Franklin Rice, Antonia Bronars, Ellande Tang, Saunon Malekshahi, Osvaldo Romero, Adrian Agogino, Alice Agogino
International Conference on Intelligent Robots and Systems (IROS), 2017
arXiv

Novel control scheme centered around simultaenous actuation of multiple cables enables fast, robust climbing on inclined surfaces.


Website source code