Profile banner
Dhiraj Gandhi

Dhiraj Gandhi

Machine learning engineer focused on robot learning and computer vision

Mountain View, California
rocketreach
Joined June 2026
Favicon imagedhiraj100892.github.io
Favicon imagegithub.com
Favicon imagelinkedin.com
Favicon imagetopmate.io

Network

565 connections

Summary

Research-focused roboticist with a track record in self-supervised learning and exploration: Dhiraj has contributed to research on self-supervised exploration and robot learning (ICML 2019 paper on exploration-by-disagreement, MSR thesis talk at CMU on self-supervised robot learning) and has co-authored work on Neural SLAM and audio-action datasets. arxiv+3
Practitioner bridging academia and industry in machine learning and computer vision: his profile lists industry roles (Google ML, Adobe internships, Barclaycard) alongside academic research positions and graduate work at Rutgers. rocketreach
Builds and releases datasets and code for reproducible robotics research: authored the Swoosh sound-action dataset and released code and project pages, and contributed to open-source implementations for Active Neural SLAM. github+2
Educational path combines VNIT undergraduate training with graduate studies and advanced coursework: holds a B.Tech from VNIT (2013–2017), completed an MS at Rutgers (2020–2021), and took advanced computer vision coursework at Princeton in 2021. rocketreach

Work

Education

Projects

Writing

Learning To Explore Using Active Neural SLAM

January 1, 2020

ICLR paper (and accompanying code) describing Active Neural SLAM, a modular approach to learning exploration policies using Neural SLAM, global and local policies.

Favicon imagegithub.com

Swoosh! Rattle! Thump! -- Actions that Sound

January 1, 2020

Paper and dataset on the interplay of action and sound in robotics, demonstrating audio's value for object identification, action prediction, and physical property inference; presented as RSS 2020 work and available on arXiv.

Favicon imagedhiraj100892.github.io

Self-Supervised Exploration via Disagreement

January 1, 2019

Paper proposing an exploration formulation training an ensemble of dynamics models and incentivizing agents to maximize ensemble disagreement to enable self-supervised exploration; demonstrated across Atari, Mujoco, Unity and implemented on a real robot.

Favicon imagearxiv.org

Hobbies

Organizing and hosting VNIT alumni events (Bay Area chapter). partiful