Nikhil Keetha (@Nik__V_

Nikhil Keetha

998 posts

Nikhil Keetha

@Nik__V__

PhD in Robotics @CMU_Robotics @airlabcmu | Visiting Researcher @Meta | Making robots 🤖 see the 🌍

Joined November 2017

Pinned
Nikhil Keetha
@Nik__V__
Sep 17, 2025
Meet MapAnything – a transformer that directly regresses factored metric 3D scene geometry (from images, calibration, poses, or depth) in an end-to-end way. No pipelines, no extra stages. Just 3D geometry & cameras, straight from any type of input, delivering new state-of-the-art
00:00
123K
Nikhil Keetha
@Nik__V__
Dec 5, 2023
SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM spla-tam.github.io We extend Gaussian Splatting to solve SLAM, i.e., automatically calculate the camera poses when fitting the Gaussian scene from RGB-D videos. Try it on your own iPhone capture today! 🧵👇
GIF
175K
Nikhil Keetha
@Nik__V__
Apr 14, 2023
Excited to share that I will be continuing at @CMU_Robotics as a PhD in Robotics student! 🤩 Thrilled to be working at the cutting edge of field robotics and perception as a part of the @AirLabCMU 🤖 #PhD #Robotics #CMU
8K
Nikhil Keetha
@Nik__V__
Mar 19, 2022
Thrilled to share that I have accepted a fully-funded Masters in Robotics offer from @SCSatCMU! Very excited to be joining the @AirLabCMU 🤖🕺 Extremely grateful to all the support from my family, friends, mentors and collaborators 🙌❤️ #gradschool #CMU #dreamcometrue #robotics
Nikhil Keetha
@Nik__V__
Aug 2, 2023
🧠🔬 Excited to share AnyLoc: Towards Universal Visual Place Recognition Foundation Models meet VPR - VPR anywhere🌍🌊🏙️, anytime🌌☁️🌄, and under anyview🚡🚗🛸 - no retraining/finetuning 🔁 - aimed at general-purpose localization & navigation anyloc.github.io 🧵👇
00:00
35K
Nikhil Keetha
@Nik__V__
Jan 31, 2024
Thrilled to share that I'll be joining Meta Zurich @AIatMeta @RealityLabs as a Summer Intern collaborating with Peter Kontschieder, @c_richardt, and many other awesome folks! Excited to push the boundaries of VR/AR, reconstruction & 🤖 scene understanding 🚀
4K
Nikhil Keetha
@Nik__V__
Mar 15, 2022
🚨 Excited to share our #CVPR2022 paper "AirObject: A Temporally Evolving Graph Embedding for Object Identification" A simple & effective temporal object encoding method. For more info: Blog: theairlab.org/airobject/ Code: github.com/Nik-V9/AirObje… Paper: arxiv.org/abs/2111.15150
GIF
Nikhil Keetha
@Nik__V__
Dec 5, 2023
Replying to @Nik__V__
We have released code for a real-time, online streaming demo, where for the first time, you can fit a Gaussian Splatting Scene live while it’s being recorded, without requiring any camera poses as input. github.com/spla-tam/SplaT… Imagine all of the uses in Robotics & AR/VR.
GIF
2.8K
Nikhil Keetha
@Nik__V__
Dec 5, 2023
Replying to @Nik__V__
Accurate camera poses are super important for Gaussian Splatting but are annoying and slow to obtain (e.g., with COLMAP). This prevents real-time applications like Robotics and VR/AR. With SplaTAM, we obtain accurate camera poses while fitting the scene with Gaussian Splatting.
GIF
3.7K
Nikhil Keetha
@Nik__V__
Dec 19, 2023
Interested in learning more about the changing landscape of robotics in the era of foundation models? Check out our latest survey & meta-analysis! Don't miss the discussion on the future of robotics 🤩🤖💡
Paper page - Toward General-Purpose Robots via Foundation Models: A Survey and Meta-Analysis
From huggingface.co
9.2K
Nikhil Keetha
@Nik__V__
Nov 30, 2023
AnyLoc has been accepted to IEEE RA-L & will be presented at @ieee_ras_icra 🥳🎉 New AnyLoc-PCA results now on Arxiv. We show that AnyLoc-PCA retains SOTA performance while having 100x smaller embedding size (~50k -> 512)
Nikhil Keetha
@Nik__V__
Aug 2, 2023
🧠🔬 Excited to share AnyLoc: Towards Universal Visual Place Recognition Foundation Models meet VPR - VPR anywhere🌍🌊🏙️, anytime🌌☁️🌄, and under anyview🚡🚗🛸 - no retraining/finetuning 🔁 - aimed at general-purpose localization & navigation anyloc.github.io 🧵👇
00:00
3K
Nikhil Keetha
@Nik__V__
Dec 5, 2023
Replying to @Nik__V__
On many difficult scenes, our approach works much better than all previous SOTA SLAM approaches. Below, we show a difficult texture-less scene where ORB-SLAM 3 completely fails, whereas ours achieves almost perfect results (first two tweets above).
GIF
2.8K
Nikhil Keetha
@Nik__V__
Feb 8, 2024
Reviewed a @CVPR paper where: Pre Rebuttal -> All Weak Reject Post Rebuttal -> All Weak Accept Kudos to the authors’ amazing rebuttal and my fellow co-reviewers! 🙌 Surprisingly all my reviewed papers have unanimous decisions 🤔😮 #CVPR2024 #R2 No more?!
10K
Nikhil Keetha
@Nik__V__
Dec 5, 2023
Replying to @Nik__V__
The key is to render a visibility silhouette from the Gaussian Scene to determine areas that have and haven’t been seen before. Using this, we can ignore unknown areas when optimizing for a new camera pose and also determine where to add new Gaussians during densification.
GIF
2.4K