Rajat Talak

About

I am a Research Scientist in the Department of Aeronautics and Astronautics at the Massachusetts Institute of Technology. I work in SPARKlab with Prof. Luca Carlone.


Prior to this, I was a Postdoctoral Associate in the Department of Aeronautics and Astronautics at MIT, from 2020-2022. I received a Ph.D. in the field of Networked Autonomy from MIT, in 2020. During this time, I worked with Prof. Eytan Modiano and Prof. Sertac Karaman in the Laboratory of Information and Decision Systems (LIDS). My work received ACM MobiHoc 2018 Best Paper Award.


For more, please see my Bio.

Research

My research interests are in the field of robot perception, certifiable models, high-level 3D scene understanding, and networked autonomy.


My current work focuses on the problems of (i) designing perception systems that are certifiable and self-supervised, (ii) 3D scene understanding with high-level scene abstractions such as 3D scene graphs, and (iii) investigating the use of language models to improve scene understanding.

You may see a complete list of my publications here or on google scholar.

News

16 Jul 2022: C-3PO: An open-source implementation of our self-supervised and certifiable pose estimation is now out [GitHub]. Paper: [arXiv]

23 Jun 2022: Our paper "Correct and Certify: A New Approach to Self-Supervised 3D-Object Perception" is now on arxiv. [arXiv]

20 Jun 2022: Gave an invited talk on Hierarchical 3D Scene Graphs and Neural Trees at the Graph Machine Learning for Visual Computing at CVPR 2022. [Workshop Link].

19 Jun 2022: Our latest work explores how we can extract common sense, from large language models, to improve robot's 3D scene understanding. Work will be presented at RSS Workshop on Scaling Robot Learning. [Workshop Link] [arXiv] (credits to: William Chen, Siyi Hu, Luca Carlone)

19-24 Nov 2021: Teaching selected lectures for the course 16.485 VNAV: Visual Navigation for Autonomous Vehicles. [MIT OCW].

18 Nov 2021: Serving as TPC member for the 5th Age of Information Workshop at INFOCOM 2022. [Workshop Page].

17 Nov 2021: Our paper on "Information Freshness in Multi-Hop Wireless Networks" is now on arXiv. [arXiv].

28 Sep 2021: Our paper on "Neural Trees for Learning on Graphs" will appear at NeurIPS 2021. [arXiv].

Contact

Email: talak@mit.edu, rajatalak@gmail.com

Address: MIT Office 31-219, 77 Massachusetts Avenue, Cambridge, MA 02139, USA