A final-year CSE Ph.D. candidate at UC, San Diego with an expected graduation date of
September, 2021 and open for job opportunities. I am interested in robotics and
artificial intelligence applied in semi-structured and dynamic environments. I am
advised by Dr. Henrik I. Christensen and Dr. Ashok K. Goel from Georgia Tech.
I am motivated by robotics applications which impact and help human counterparts in
the real-world. Given this motivation I find model-based reasoning a strong tool to
frame planning problems thereby making them interpretable and introspectable. This
knowledge can then be used to generate dialogues between a human and a robot to either
guide the robot for a novel problem or explain the reason for departing from a plan or
why a failure occurred. My research-work focuses on approaches that use existing
structure to make data-based learning more efficient and enable its interpretable
explanation for collaborative robotics.
My research has been looking at how to ground an abstract action like
Pick('pulley') along-with its preconditions and effects into physical
variables and observations so that a robot can monitor progress of its own plan. If
the robot fails, we use a trace of this plan and its grounding to reason what went
wrong and learn its solution. My thesis is about doing this kind of failure repair at
multiple levels of planning, i.e. skill, task and mission specifications. In
manufacturing we can use these grounded explanations along with high-level knowledge
of domain to create rich set of interactions between a manufacturing robot and its
Apart from academia, I have also interned for Vecna Robotics and Honda Research
Institute - USA. While the earlier years of my Masters and Ph.D. research focused on
mobile platforms and navigational problems in dynamic environments, my thesis grounds
model-based reasoning for application on manipulators.
My projects can be accessed from the sidebar, or you can checkout the following stream
of dated posts:
Our planning paper outlining the architecture, planning stack, and problem &
domain formulation for the World Robot Summit’s Assembly Challenge got accepted at CASE 2021!
A pre-print can be found at https://arxiv.org/abs/2103.07544.
The World Robotics Challenge (2018 & 2020) was designed to challenge teams to design systems that are easy to adapt to new tasks and to ensure robust operation in a semi-structured environment. We present a layered strategy to transform missions into tasks and actions and provide a set of strategies to address simple and complex failures. We propose a model for characterizing failures using this model and discuss repairs. Simple failures are by far the most common in our WRC system and we also present how we repaired them.
We spent a good amount of last year building a dual-arm system capable of assembling a
product made of known parts and of known/defined design. This work was funded by RPDC
and we developed the system under the name “RPDC Robotics” for the 2021 Assembly
Challenge at World Robot Summit 2020 (now postponed to Summer 2021). You can check out
our reports on Arxiv:
- System and Experiences - https://arxiv.org/abs/2103.15236
- Reasoning and plan execution - https://arxiv.org/abs/2103.07544
- Vision for manufacturing domain - https://arxiv.org/abs/2011.00372
Videos showing our vision-based manipulation:
Here’s a cool video showing off a dual-arm assembly task: