Devendra Singh Chaplot is an Indian-origin artificial intelligence research scientist known for his work in robotics, embodied AI, deep reinforcement learning, and large language models (LLMs). An alumnus of the Indian Institute of Technology Bombay and Carnegie Mellon University, Chaplot has held key roles at major technology companies and AI labs, including Meta AI, Mistral AI, and SpaceX/xAI. He gained widespread media attention in March 2026 when he joined Elon Musk's initiatives to develop artificial superintelligence. ^{[1] [2]}​

Education

Chaplot completed his undergraduate studies at the Indian Institute of Technology Bombay (IIT Bombay), earning a Bachelor of Technology (B.Tech) in Computer Science & Engineering with a minor in Applied Statistics in August 2014. ^{[3] [4]}

He then pursued his graduate studies in the United States at Carnegie Mellon University (CMU) in the School of Computer Science. He began his doctoral studies in August 2016 and received both a Master of Science (M.S.) and a Ph.D. in Machine Learning. His doctoral work culminated in a thesis defended in March 2021 titled "Building Intelligent Autonomous Navigation Agents." His Ph.D. advisors were professors Ruslan Salakhutdinov and Abhinav Gupta. ^[3]

Career

Chaplot's career has spanned internships and full-time research positions at several major technology and AI firms. During his undergraduate and graduate studies, he held several roles, including as a Research Associate at Samsung Electronics in Suwon, South Korea, from September 2014 to August 2015 in the IoT Intelligence Lab. While pursuing his Ph.D., he worked as a researcher at Apple AI Research from June 2017 to December 2017 and as a Research Intern at Facebook AI Research (FAIR) from May 2018 to December 2020. ^{[1] [3]}​

Meta AI (FAIR)

Chaplot joined Facebook AI Research (FAIR), later rebranded as Meta AI, as a Research Scientist. His tenure lasted from December 2020 to August 2023 in Menlo Park, California. At FAIR, his research focused on the intersection of computer vision and robotics, building on the work he began during his graduate internships and doctoral studies. ^{[1] [3]}​

Mistral AI

In April 2023, Chaplot became a founding team member of Mistral AI, a Paris-based startup that quickly gained prominence in the field of large language models. During his time there, which lasted until July 2024, he made significant contributions to the company's foundational models. He was involved in the training of models such as Mistral 7B, Mixtral 8x7B, and Mistral Large. Chaplot also led the multimodal research team, which developed the Pixtral 12B and Pixtral Large models. Additionally, he was responsible for establishing and leading Mistral AI's U.S. office in Palo Alto, California. ^{[1] [3]}​

Thinking Machines Lab

According to his professional profile, Chaplot was a founding team member at Thinking Machines Lab from February 2025 to March 2026. In his research and product role, he contributed to the development of "Tinker," an API designed for users to train their own Large Language Models. ^[3]​

xAI and SpaceX

In March 2026, Chaplot announced via a post on X (formerly Twitter) that he was joining SpaceX and xAI. The move attracted significant media coverage in both Indian and international technology press. His stated goal was to "work with Elon Musk and his team on building superintelligence to help humanity." ^{[1] [2]}​

Research and Key Projects

Chaplot's research primarily focuses on creating intelligent agents that can perceive, navigate, and interact with complex 3D environments, both simulated and physical. His work often combines deep reinforcement learning with computer vision and classical robotics techniques. ^{[1] [3]}

Active Neural SLAM

One of Chaplot's notable projects is Active Neural SLAM. This work integrates deep reinforcement learning with classical Simultaneous Localization and Mapping (SLAM) methodologies to create autonomous exploration agents. The system learns to navigate efficiently in unfamiliar environments. This project was recognized as the winner of the CVPR 2019 AI Habitat Navigation Challenge. ^[3]

Object Goal Navigation

Chaplot also worked on Object Goal Navigation ("ObjectNav"), a foundational task in embodied AI where an agent is required to locate a specific object (e.g., "find a chair") in a previously unseen environment. His framework enabled agents to perform this task with high success rates, and the research garnered attention from technology publications such as TechCrunch and Gizmodo. His work in this area won the CVPR 2020 Habitat ObjectNav Challenge. ^[3]

"Arnold" Doom AI Agent

In 2017, Chaplot developed "Arnold," an autonomous AI agent designed to play the classic first-person shooter game Doom. The agent demonstrated advanced capabilities by learning to navigate complex, dynamic 3D levels and engage in combat with enemies from visual input alone. The "Arnold" agent achieved first place in the Full Deathmatch track of the Visual Doom AI Competition in 2017 and also received the Best Demo Award at the AAAI-17 conference. ^[3]

PyRobot

While at Facebook AI Research, Chaplot was involved with PyRobot, an open-source robotics framework designed to democratize AI and robotics research. PyRobot provides a lightweight, high-level interface over a robot's low-level controllers, making it easier for researchers to run experiments and benchmark performance across different hardware platforms. This project received a Facebook Research Award in October 2019. ^[3]

Awards and Honors

Throughout his academic and professional career, Devendra Chaplot has received several awards and fellowships in recognition of his research contributions.

• Facebook Graduate Fellowship (January 2020)
• Nvidia Graduate Fellowship (December 2019) - Awarded, but declined by Chaplot.
• Facebook Research Award (October 2019) - For the project "PyRobot: Democratizing Robotics."
• Winner, Habitat ObjectNav Challenge (CVPR 2020)
• Winner, Habitat Navigation Challenge (CVPR 2019)
• Best Paper Award (CVPR-18 Deep Learning for Visual SLAM workshop, 2018)
• First Place, Full Deathmatch (Visual Doom AI Competition, 2017) - For the "Arnold" agent.
• Best Demo Award (AAAI-17, 2017) - For the "Arnold" agent.^[3]