Shruti Joshi

shrutijoshi98 at gmail dot com, shruti dot joshi at tuebingen dot mpg dot de

I am a software programmer/research intern at the Empirical Inference Department at MPI-IS, Tübingen, being advised by Prof. Bernhard Schölkopf, Stefan Bauer, and Manuel Wüthrich.

I graduted from the Department of Electrical Engineering at the Indian Institute of Technology Kanpur (IIT Kanpur) in May, 2019. Over the course of my undergraduate studies, I have been lucky to have worked with some great advisors. I worked under the guidance of Prof. Dmitry Berenson at the University of Michigan, Ann Arbor, on a research problem which addressed the difficulties encountered in the manipulation of a highly deformable contact manifold such as paper. At IIT Kanpur, I have worked with Prof. Indranil Saha on multi-agent pursuit evasion in a patrol workspace, and with Prof. Ketan Rajawat on energy efficient online trajectory optimization.

Email  /  Github  /  Google Scholar  /  Twitter


  • New pre-print out: FCRL! We investigate how well we can generalize to a bunch of different downstream tasks by attempting to learn a generic, task-agnostic representation of the underlying data-generating process itself.
  • Our paper introducing the TriFinger robots has been accepted at CoRL'20!
  • The Real Robot Challenge has been launched! Follow us on Twitter for the latest updates!


My research interests are focused primarily on reinforcement learning, deep learning, and causality.

How can we (sample)efficiently learn decision making policies on diverse, open-world data? What kind of world models provide meaningful representations about objects and other entities to an agent interating with its environment? What causal information can we already incorporate towards solving the problem of credit assignment, and how do we uncover the underlying causal structure of the decision-making problem? These are some of the questions that motivate my current research.

I am also interested in better software design to reliably test out my research hypotheses, and I continue to learn how to build well-maintained (!) iterative codebases and how to better debug and visualize my work, not letting pesky bugs steal my motivation.

TriFinger Simulation
Shruti Joshi, Felix Widmaier, Vaibhav Agrawal, Manuel Wüthrich. 2020  
This package provides a simulation interface to the TriFinger robots.
Function Contrastive Learning of Transferable Representations
Muhammad Waleed Gondal, Shruti Joshi, Nasim Rahaman, Stefan Bauer, Manuel Wüthrich, Bernhard Schölkopf. In submission
We propose a self-supervised contrastive learning method that meta learns a function representation driven by the self-supervision signal indicating whether two sets of samples stem from the same underlying function.
TriFinger: An Open-Source Robot for Learning Dexterity
Manuel Wüthrich, Felix Widmaier, Felix Grimminger, Joel Akpo, Shruti Joshi, Vaibhav Agrawal, Bilal Hammoud, Majid Khadiv, Miroslav Bogdanovic, Vincent Berenz, Julian Viereck, Maximilien Naveau, Ludovic Righetti, Bernhard Schölkopf, Stefan Bauer. CoRL'20
We present a novel open-source, low-cost, robotic platform for learning dexterous object manipulation.
Online Utility-Optimal Trajectory Design for Time-Varying Ocean Environments
N Mohan Krishna, Shruti Joshi, Ketan Rajawat. ICRA, 2019  
We showed that a completely online planner could successfully navigate to a mobile goal location in a simulated ocean environment while keeping energy expenditure to a minimum. code
Explanations for Temporal Recommendations
Homanga Bharadhwaj, Shruti Joshi. XAI Workshop, IJCAI/ECAI, 2018  
We show that a neighborhood style explanation scheme can be used as an auxiliary mechanism for interpreting the predictions of a Recurrent Neural Network based temporal recommendation model.
I also contribute to the Open Dynamic Robot Initiative. Some of my contributions include the development of a realtime logging interface for the TriFinger robots, and drivers for image acquisition from the cameras on the TriFinger platforms.

I love his website design.