Picking with Purpose: Task-driven and Observable Manipulation
@ The Mcube Lab, MIT                                        December 2019 - Present
Research Advisor: Prof. Alberto Rodriguez, MIT
We are building a framework to train robots to learn to selectively pick up objects directly from a point cloud image for better observability and manipulability for a given task. Currenly, we are exploring the application of this framework to industrial kitting operations where a robot needs to pick up objects from a pile and place them accurately in specific configurations.
Stay tuned for a paper and videos!
Prehensile Pushing: Planning and Control for In-Hand Manipulations Using External Contacts
@ The Mcube Lab, MIT                                        September 2014 - December 2019
Research Advisor: Prof. Alberto Rodriguez, MIT
       I am empowering robots with a reasoning to manipulate an object in grasp with a series of external pushes. The external pushes abstract the interactions with the environment, a second robot arm or extra fingers of a multi-finger gripper. Our continuous development on this project has lead to a framework which can:
1) simulate - predict how the object is going to move in the grasp when pushed and what forces-and-torques we should expect to observe at all the contacts.
2) plan - reason about long-horizon pushing strategies, potentially involving pushes from different sides of the object, to achieve a desired grasp on the object.
3) control - handel uncertainty in system parameters and external disturbances by generating local corrective pushes in MPC fashion to track the planned strategies.
      With our fast planning and control framework, which generates regrasp strategies in less than a second and executes them robustly, we are looking at broader problems such as assembly planning and efficient pick-and-place/pack/use.
7] Planar In-Hand Manipulation via Motion Cones, IJRR 2019 (Invited paper) [Paper]
6] Regrasping by Fixturelss Fixturing, CASE 2018 [Paper]
5] In-Hand Manipuation via Motion Cones, RSS 2018 (Best Student Paper award) [Paper]
4] Stable Prehensile Pushing: In-Hand Manipulation with Alternating Sticking Contacts, ICRA 2018 [Paper]
3] Sampling-based Planning of In-Hand Manipulation with External Pushes, ISRR 2017 [Paper]
2] Experimental Validation of Contact Dynamics for In-hand Manipualtion. ISER 2016 [Paper]
1] Prehensile Pushing: In-Hand Manipulation with Push Primitives, IROS 2015 [Paper]
Featured on MIT homepage  [link] --- YouTube Video by MIT [link] --- MIT news article [link] --- TechCrunch article [link]
Design for Dexterity : An Algorithmic Design of Hand-Robot-Environment Systems for Dexterous Manipulation
@ The Mcube Lab, MIT             January 2014 - Present
Research Advisor: Prof. Alberto Rodriguez, MIT
We propose simultaneous design of arm motions, shapes of phalanxes of a robot hand and features in the environment around a robot to facilitate fast and reliarble in-hand manipulation of various objects. The algorithmic design of these elements - Design for Dexterity - will make  dexterous manipulation practical for robots for industrial assembly operations or field applications.
Robotic systems designed for dexterity can simplify manipulation planning in uncertain environments. As such capability will be derived from a complete hand-robot-environment system rather than from only the hand, it will be be accessible to a variety of robot hands, industrial manipulators and robots. 
Extrinsic Dexterity: In-Hand Manipulation using External Forces
@ The Manipulation Lab, CMU        August 2012 - December 2013
@ The Mcube Lab, MIT            January 2014 - Present
Research Advisors: Prof. Matt Mason, CMU and Prof. Alberto Rodriguez, MIT
This research is a part of Simple Hands project which aims to develop a simple gripper, along with control and planning techniques, for general purpose manipulation.
The simple hand designed at the Manipulation Lab, called the MLab Hand, has three fingers which are compliantly coupled to a single actuator. The hand is attached to an industrial robot arm, ABB IRB 140. The design simplicity of the MLab hand limits its intrinsic dexterity. However, we believe, the manipulation dexterity of a robot is not limited to the intrinsic dexterity of a hand; we can exploit external resources such as gravity, arm motions, object inertia and external contacts to facilitate the dexterous manipulation of objects.
We have developed fourteen regrasp actions which change the pose of an object in the hand. Repertoire of these regrasp actions allows the robot to transfer an object from one pose to another using a sequence of regrasp actions if not possible in a single move.
Each grasp graph shown below represents an entire repertoire of regrasp actions, showing the grasp type transitions, from which it is easy to identify reachable grasp types and possible paths through the grasp types.
                                                                                                         Fig : Grasp Graph for triangular prism 
                                                                                                             Fig : Grasp Graph for cylindrical objects
One of the goals for this projects is to understand the transition mapping between these regrasps and develop a learning-based framework to plan a sequence regrasp actions to transfer the object from initial configuration to the desired one. 
Extrinsic Dexterity: In-Hand Manipulation with External Forces [Paper]
Media: (ICRA 2014 Finalist for Best Research Video)
2] MIT news article [link]
Finger Design for facilitating specific regrasp actions
@ The Manipulation Lab, CMU        May 2013 - December 2013
@ The Mcube Lab,  MIT            January 2014 - Present
Research Advisor: Prof. Matt Mason   and Prof. Alberto Rodriguez, MIT
I designed fingertips for a simple parallel-jaw gripper to enable it to perform a specific but very demanding regrasp action of pivoting a prismatic object while grasping.
This work is being pursued for commercial applications [United States Patent Application 20170036354].
2] Pneumatic Shape-shifting Fingers to Reorient and Grasp [Paper]
1] A Two-Phase Gripper to Reorient and Grasp [Paper]