Somnath Kumar

Research Engineer @ Scaled Foundations.

avatar.jpeg

I am a research engineer at Scaled Foundations, working on Data generation for General Robotics, Previously, I was a Research Fellow at Microsoft Research focusing on alignment and optimization of language models using Reinforcement learning. My interests encompass but not limited to Representation learning and Embodied AI. My research work has been interpreting and developing data-efficient systems for General Robotics.

In my past life I had completed my undergrad at Indian Institute of Technology (BHU), Varanasi. While, I worked with Qin Lin and John Dolan at Argo AI Lab at Robotics Institute, Carnegie Mellon University working on Safe Autonomous Navigation. Prior to that, I interned under Dr. Shishir Kolathaya, IISc, bangalore at Stochlab on Efficient Reinforcement Learning for Quadrupeds.

For more details download my resumé

news

Mar 2025 Our paper VidModEx has been published in the proceedings of the 5th Adversarial Machine Learning Workshop at CVPR 2025.
Nov 2024 Our paper Bridging the Gap is published at COLING 2025.
Sep 2024 Our paper MMCTAgent has been accepted to the Open World Agent Workshop at NeurIPS 2024.
Aug 2024 I’ve joined Scaled Foundations to work on general robotics intelligence. Excited for what lies ahead!
Feb 2024 Our work Shiksha has been highlighted and covered by Satya Nadella at various venues.
Oct 2023 Our Project Shiksha won first place in Microsoft Hackathon’23 in both national region and worldwide.
Jul 2023 Started working as a Research Fellow in Microsoft Research India
Nov 2022 Appointed as the Deputy Contingent Leader representing my Institution for the 11th InterIIT Techmeet.
May 2022 Joined as a Summer Intern at AI Shield Department of Bosch, Bangalore
Mar 2022 Bagged Silver in Bosch Model Extraction Event of Inter IIT Tech meet 2022 against 22 IIT’s.
Dec 2021 Led the team which was First Runners in AIITRA Robotics Challenge.
Aug 2021 Appointed as the Joint Secretary of Club of Programmers, IIT Varanasi.

selected papers

  1. vidmodex_main_webpage.svg
    VidModEx: Interpretable and Efficient Black Box Model Extraction for High-Dimensional Spaces
    Somnath Sendhil Kumar, Yuvaraj Govindarajulu, Pavan Kulkarni, and 1 more author
    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, Jun 2025
    PDF Code
  2. mmct.png
    MMCTAgent: Multi-modal Critical Thinking Agent Framework for Complex Visual Reasoning
    Somnath Kumar, Yash Gadhia, Tanuja Ganu, and 1 more author
    In Workshop on Open-World Agents: Synnergizing Reasoning and Decision-Making in Open-World Environments (OWA), Nov 2024
    arXiv Code
  3. leap.png
    Bridging the Gap: Dynamic Learning Strategies for Improving Multilingual Performance in LLMs
    Somnath Kumar, Vaibhav Balloli, Mercy Ranjit, and 5 more authors
    In Proceedings of the 31st International Conference on Computational Linguistics, Nov 2024
    PDF Code
  4. latentcbf.png
    LatentCBF: A Control Barrier Function in Latent Space for Safe Control
    Somnath Sendhil Kumar, Qin Lin, and John Dolan
    Nov 2023
    PDF
  5. stochlite.png
    Realizing Linear Controllers for Quadruped Robots on Planetary Terrains
    Aditya Shirwatkar, Somnath Sendhil Kumar, Shamrao Garur, and 2 more authors
    In Proceedings of the 2023 6th International Conference on Advances in Robotics, Mar 2023
    PDF
  6. bgaitrnet.png
    BGaitR-Net: An effective neural model for occlusion reconstruction in gait sequences by exploiting the key pose information
    Somnath Sendhil Kumar, Binit Singh, Pratik Chattopadhyay, and 2 more authors
    Expert Systems with Applications, Mar 2024
    PDF Code