My primary research objective is to enable reactive robotic manipulation, moving towards robots that can better perceive their environment and react to unforeseen situations. My research aims to enable robots skilled at i) monitoring the execution of a manipulation task, and ii) using sensed information in real-time for closed-loop control. I am particularly interested in developing reactive manipulation policies relying on visual and tactile feedback for contact-rich robotic manipulation tasks where the dynamics are dominated by frictional interactions.

Tactile Dexterity: Manipulation Primitives with Tactile Feedback

Tactile

Tactile dexterity is an approach to dexterous manipulation that plans for robot/object interactions that render interpretable tactile information for control. We develop closed-loop tactile controllers for dexterous manipulation with dual-arm robotic palms with the ability to manipulate an object from an initial pose to a target pose on a flat surface while correcting for external perturbations and uncertainty in the initial pose of the object. We validate the approach with an ABB YuMi dual-arm robot and demonstrate the ability of the tactile controller to handle external perturbations. Key to this formulation is the decomposition of manipulation plans into sequences of manipulation primitives with simple mechanics and efficient planners.

Project Page


Hybrid Feedback Control For Contact Interactions

Tactile

We investigate investigates real-time control strategies for dynamical systems that involve frictional contact interactions. Hybridness and underactuation are key characteristics of these systems that complicate the design of feedback controllers. Our results show that a Model Predictive Control approach used in tandem with integer programming offers a powerful solution to capture the dynamic constraints associated with the friction cone as well as the hybrid nature of contact.

WAFR 2016 ICRA 2018


Data-Efficient Control for Planar Manipulation

Tactile

This paper explores the data-complexity required to control manipulation tasks with a model-based approach, where the model is learned from data. We employ this methodology to the problem of pushing an object on a planar surface, and find that we can design effective control policies with small data requirements (less than $10$ datapoints) while achieving accurate closed-loop performance. </p>

CoRL 2018


Autonomous Grasping

ARC

During the period 2015-2017, I participated in the Amazon Robotics Challenge (ARC) as part of team MIT-Princeton. The challenge involved developing a robotic system capable of recognizing and grasping novel objects in cluttered environments. Our approach made use of an object agnostic deep learning framework that directly maps visual observations to planned robot motion. This work has led to a 1st place in the 2017 stowing task challenge as well as a 2018 Amazon Best System Paper Award. </p>

ICRA 2018


Tactile Regrasping

Tactile

In this project, we develop tactile reflexes, giving robots the ability to make grasp adjustments immediately upon making contact with an object. Tactile information is used to assess in real-time the quality of a grasp and predict failures preemptively and to design a controller able to perform local grasp adjustments, where grasp are improved directly upon making initial touch with an object. </p>

IROS 2018