Nishad Gothoskar

I am a second-year PhD Student at MIT EECS advised by Vikash Mansinghka and Josh Tenenbaum. I am a member of the Probabilistic Computing Project in CSAIL and BCS. Previously I was a researcher at Vicarious AI where I worked with Miguel Lázaro-Gredilla and Dileep George, and engineer at Uber ATG. In Dec. 2017, I recieved a BS in Computer Science and Math from Carnegie Mellon.

My research aims to build robots that can learn as rapidly and efficiently and generalize as broadly as humans. Humans have rich prior knowledge and inductive biases about the structure of the world, and they leverage this understanding when making inferences from data. To build AI systems as flexible as humans, we must understand what these priors and biases are, how our brains represent them, and how we use them online. In my research, I apply probabilistic generative models to improve data-efficiency, robustness, and generalizability of robotic systems.

Contact me at nishad AT mit DOT edu

[Google Scholar] [LinkedIn]

DURableVS: Data-efficient Unsupervised Recalibrating Visual Servoing via online learning in a structured generative model
Nishad Gothoskar, Miguel Lazaro-Gredilla, Yasemin Bekiroglu, Abhishek Agarwal, Joshua B. Tenenbaum, Vikash K. Mansinghka, Dileep George
ICRA, 2022 [PDF]

In this work, we present a method for unsupervised learning of visual servoing that does not require any prior calibration and is extremely data-efficient. Our key insight is that visual servoing does not depend on identifying the veridical kinematic and camera parameters, but instead only on an accurate generative model of image feature observations from the joint positions of the robot. We demonstrate that with our model architecture and learning algorithm, we can consistently learn accurate models from less than 50 training samples (which amounts to less than 1 min of unsupervised data collection), and that such data-efficient learning is not possible with standard neural architectures. Further, we show that by using the generative model in the loop and learning online, we can enable a robotic system to recover from calibration errors and to detect and quickly adapt to possibly unexpected changes in the robot-camera system (e.g. bumped camera, new objects).

3DP3: 3D Scene Perception via Probabilistic Programs
Nishad Gothoskar, Marco Cusumano-Towner, Ben Zinberg, Matin Ghavamizadeh, Falk Pollok, Austin Garrett, Dan Gutfreund, Joshua B. Tenenbaum, Vikash Mansinghka
NeurIPS, 2021 [MIT News] [PDF]

We propose a generative probabilistic programming-based architecture for modeling 3D objects and scenes, and use our architecture to do accurate and robust object pose estimation from RGBD images.

Clone-structured graph representations enable flexible learning and vicarious evaluation of cognitive maps
Dileep George, Rajeev V. Rikhye, Nishad Gothoskar,J. Swaroop Guntupalli, Antoine Dedieu, Miguel Lazaro-Gredilla
Nature Communications, 2021 [PDF]

Cognitive maps are mental representations of spatial and conceptual relationships in an environment, and are critical for flexible behavior. To form these abstract maps, the hippocampus has to learn to separate or merge aliased observations appropriately in different contexts in a manner that enables generalization and efficient planning. Here we propose a specific higher-order graph structure, clone-structured cognitive graph (CSCG), which forms clones of an observation for different contexts as a representation that addresses these problems.

Query Training: Learning a Worse Model to Infer Better Marginals in Undirected Graphical Models with Hidden Variables
Miguel Lazaro-Gredilla, Wolfgang Lehrach, Nishad Gothoskar, Guangyao Zhou, Antoine Dedieu, Dileep George
AAAI, 2021 [PDF]

Probabilistic graphical models (PGMs) provide a compact representation of knowledge that can be queried in a flexible way: after learning the parameters of a graphical model once, new probabilistic queries can be answered at test time without retraining. However, when using undirected PGMS with hidden variables, two sources of error typically compound in all but the simplest models (a) learning error (both computing the partition function and integrating out the hidden variables is intractable); and (b) prediction error (exact inference is also intractable). Here we introduce query training (QT), a mechanism to learn a PGM that is optimized for the approximate inference algorithm that will be paired with it.

Template from here.