Activity
1K followers
-
Xiaqing Pan shared this🚨🚨🚨 We Are Hiring!!! My team is hiring a new-grad PhD Research Engineer with robotics background. If you are interested, please apply for the job at (please click this link directly instead of the link preview) https://lnkd.in/dWtzFnx2 This role will be based in Redmond. You will have chance to solve cutting edge problems including but not limited to navigation, dex-hand manipulation, whole-body controlling and long horizon task reasoning. More exciting, you will intersect the robotics work with the world-leading egocentric research foundation that we have built for many years - Project Aria (projectaria.com)
-
Xiaqing Pan liked thisXiaqing Pan liked thisOn May 29th we are officially opening our Flagship 3D Tech and 3D Scanning Center in Brooklyn, NYC. And we do so with a BANG. 🔥 ✅ Super curated event with entrepreneurs, investors and builders in the Generative 3D and Physical AI space. ✅ Panel Discussion with our customers Nike and Amazon on "generative 3D AI" ✅ Italian-American Aperitivo, Food and Music 🍕 🍷 🗽 There are a few spots left for my #NYC connections. Please PM me in the next days if you are interested in joining. 👊
-
Xiaqing Pan liked thisXiaqing Pan liked thisHuman data becomes far more useful when robots are human-like. Excited to share a milestone from our work with GEAR at NVIDIA: we trained a VLA model on 20k+ hours of in-the-wild human data and deployed it on robots with 22-DoF hands. Key findings: - Near log-linear scaling between human data volume and action accuracy (R² = 0.998), predictive of real dexterous performance - Few-shot generalization begins to emerge at this scale, with some tasks solved from a single demo - Policies trained on humans transfer across embodiments, including lower-DoF hands Simple recipe + scale = new capabilities We also discovered other emerging properties such as strong language following. More exciting stuff to come! Website: https://lnkd.in/eVk48GRj Paper: https://lnkd.in/eW4bbbQF
-
Xiaqing Pan liked thisXiaqing Pan liked thisHappy to share that I have recently joined the GEAR team at NVIDIA to lead a new effort on scaling humanoids with human data. We are building the human-to-robot data flywheel that makes generalist robots inevitable. 🚀 We’re hiring across the board. If this vision excites you, reach out.
-
Xiaqing Pan liked thisXiaqing Pan liked thisExcited to announce that I’m holding interviews for a 2026 PhD Intern! 🎉 If you know me, you know my passion for applied social science and how it drives my work every day at Meta. I’m eager to connect with talented individuals who share this enthusiasm and are ready to make an impact. If you or someone you know is interested in joining a team that values research, collaboration, and real-world impact, let’s connect! Please share this post to help spread the word. I am particularly interested in speaking to students with an interest in heterophenomology and achievements in intersubjectivity. #Hiring #AppliedSocialScience #UXResearch #JoinUs https://lnkd.in/gdAbqfvGUX Research Scientist Intern, Multimodal Conversation Analyst (PhD)UX Research Scientist Intern, Multimodal Conversation Analyst (PhD)
-
Xiaqing Pan liked thisWe're developing the future of human-robot interaction. Very excited to be part of this ground breaking team.Xiaqing Pan liked thisUnveiling agentic robotics on GRID –– our blueprint for machines that can reason, converse, compose, and remember. Agents have transformed software, now it's time for robotics. – Modular skills (segmentation, grasping, planning) that can be swapped in and out, and hosted on remote infrastructure for maximum accessibility. – Robots as tools where heterogeneous embodiments are exposed through a common API layer, not locked behind fragmented SDKs. – Composition & adaptation: agents chaining skills into new behaviors — composing when skills exist, and creating when they do not by iterating autonomously with simulators. – Memory: Vector databases for robotics enabling agents to understand and remember their experiences as well as operating procedures. This vision is already here. We’ve seen GRID: ⚡️ Build robot manipulation skills and use them for different objects and varying tasks. ⚡️ Play robot chess through board recognition → motion planning → reasoning. ⚡️ Run across embodiments — humanoids, quadrupeds, drones — with the same agentic framework. The era of general-purpose robots won’t be defined by a monolithic model. It will be defined by agentic architectures that reason, compose, converse, and remember. Read more in our blog: https://lnkd.in/eHAKbem3 Paper: https://lnkd.in/eZeudC5y Credits to the incredible team that contributed to this effort: Sai Vemprala, Griffin Norris, Brandon Rishi, Pranay Reddy, Sai Satvik Vuppala, Arthur Lovekin, Mayank Garg, Pushkal Katara, Tianshuang Qiu, Nathaniel Chin, Preet Shah, Saif Punjwani, Sarthak Bhagat, Jonathan Huang, Shuhang Chen, Ashish Kapoor
-
Xiaqing Pan liked thisETH Zurich's LaMAria Dataset will be presented as a highlight paper at #ICCV2025 next week in Honolulu. Project Aria (https://lnkd.in/g8Tjb48P) was used to record ~22 hours and 70km of movement through a city center, with Meta's SLAM algorithms computing closed-loop and bundle-adjusted trajectories on all of the recordings. This breakthrough city-scale centimeter-accurate benchmarking dataset with ground truth will help accelerate the state of the art in SLAM algorithm development for years to come. Visit the LaMAria tutorial at ICCV on Oct 19th to learn more: https://lnkd.in/gM7ZzTFN. See the paper and code links in the original post.Xiaqing Pan liked thisWe've released our latest work, to be presented as Highlight at #ICCV2025! We're introducing a new, large-scale dataset to push the boundaries of egocentric visual-inertial Simultaneous Localization and Mapping (SLAM). Modern SLAM systems are still brittle when faced with the challenges of real-world egocentric data (think AR/VR glasses). To measure subtle effects like metric scale drift down to 0.1%, we needed long trajectories and extremely accurate ground truth. This is what we set out to create with LaMAria. Our new benchmark features: > 🌆 City-scale trajectories: ~22 hours and 70km of data recorded with Aria glasses. > 🎯 Centimeter-accurate ground truth: We used survey-grade instruments to create a network of control points, ensuring our ground truth is orders of magnitude more accurate than the best existing systems. > ☠️ Extreme challenges: We have it all! Long trajectories (up to 45 mins!) with time-varying sensor calibrations, fast egocentric motions, within moving environments (a tram!) or extreme low-light conditions. We found that even state-of-the-art open-source SLAM systems struggle on our data. We even had to create an extra "easy" dataset just to show that our evaluation setup was correct! This reveals a significant gap and a huge opportunity for the community to develop the next generation of robust, multi-sensor, self-calibrating, and scalable SLAM systems. To spur further research, we are releasing all our data and code with permissive licenses. We also provide an evaluation server with a public leaderboard to track progress. As a bonus, we are contributing experimental support for visual-inertial optimization in COLMAP. We hope LaMAria will be a valuable resource for the community. Come and join our tutorial at ICCV to learn more about the dataset and how to use it! Paper: https://lnkd.in/dENB43t2 Website: https://lamaria.ethz.ch Code: https://lnkd.in/dTwK-h2Z Tutorial: https://lnkd.in/dWSnPmdz This was a huge team effort led by amazing Anusha Krishnan and Shaohui Liu at ETH Zürich with Oscar Gentilhomme & Marc Pollefeys in collaboration with David Caruso, Maurizio Monge, Richard Newcombe, and Jakob Engel at Meta Reality Labs, who provided the devices and funding to make this project possible. Thank you all!
-
Xiaqing Pan liked thisXiaqing Pan liked thisA fun perk of attending the Samsung AI award event was meeting Yoshua Bengio and having a fun debate on the future of humanity with super-intelligent robots. If you ask me, the future looks bright -- there has never been a greater need for skilled labor that humanity shouldn’t carry alone.
-
Xiaqing Pan liked thisXiaqing Pan liked this7am in the factory and another Sphere 3D scanner is all packed up & ready for shipment. NYC bound. 🇺🇸🗽 Made in Italy. 🇮🇹 What a team.
Experience
Education
-
University of Southern California
3.88/4.0
-
Dissertation title: Machine Learning Methods for 2D/3D Shape Retrieval and Classification
Advisor: C.-C. Jay Kuo -
-
-
Publications
-
A two-stage shape retrieval (TSR) method with global and local features
Journal of Visual Communication and Image Representation
Projects
-
Virtual View Synthesize for Movie Quality 2D to 3D conversion
The project is a joint attempt with a movie special effect studio. The objective is to improve the state-of-the-art 3D movie production pipeline. The proposed improvements including creating highly immersive 3D visual experience, and efficiently reducing human involvement in the algorithm. The algorithm has already been applied in actual movie production. We are currently working on a much more advanced upgrade to the current algorithm.
Other creators
Honors & Awards
-
MS Student Honors Program
Ming Hsieh Dept. Of Electrical Engineering (USC)
The MSHP is a highly selective program that offers outstanding students the chance to:
1. Network with faculty members and PhD students
2. Meet with key Ming Hsieh Department industrial partners
3. Have special access to Viterbi Career Services resources
4. Receive invitations to MSHP events
5. Be granted priority for funding/research opportunities, including:
6. Directed Research and/or CPT (curricular practical training) supervision
Languages
-
English
Full professional proficiency
-
Chinese
Native or bilingual proficiency
-
Shanghainese
Native or bilingual proficiency
Other similar profiles
Explore more posts
-
ZETIC
On-device AI becomes powerful only when it works reliably on real hardware. In this Melange Walkthrough, Yeonseok Kim and Sun-Q Kim break down how the full on-device pipeline works, from model selection and quantization to real-device benchmarking and SDK deployment. Select. Benchmark. Deploy. This is how production-ready on-device AI should operate. Watch the full walkthrough below. #OnDeviceAI #EdgeAI #MobileAI #AIInfrastructure #ZETIC #Melange
-
Bobby Napiltonia
Rahul Kar and I just spent the week at NVIDIA GTC, and if you missed Jensen Huang keynote, here’s the real takeaway no one is saying loudly enough: AI is no longer constrained by models. It’s constrained by power ⚡ ⚡ We’ve officially crossed the line where #compute is abundant, capital is flowing, demand is real… but none of it matters if you can’t actually turn the #GPUs on. Let that sink in................................. The next phase of #AI isn’t about #training bigger models. It’s about running them. #Inference is the #economy. And inference runs on #power, not hype. What’s clear is this: the bottleneck has shifted. It’s NOW MEGAWATTS. And here’s the part most people are still missing…....... We don’t have a pure supply problem. We have an orchestration problem. There is #strandedpower sitting inside existing data centers right now. Underutilized. Fragmented. Locked behind legacy assumptions. While everyone is out chasing new builds that take years, the fastest path forward is unlocking what’s already there. TODAYS POWER can support TOMORROWS AI This is the game: Maximize the power envelope Turn stranded capacity into AI-ready infrastructure Collapse deployment timelines from years to months That’s how you win the inference economy!!!!! At HammerheadAI, Inc., this is exactly what we’re focused on. Not just finding more power, but using what already exists in a radically more intelligent way. Because the companies that figure this out first won’t just run AI…They’ll define the economics of it. And it’s 🔥 🔥 ❤️🔥 If you're a #CRO, infra leader, or operator and you're still thinking in terms of GPUs instead of megawatts, you're already behind. This shift is happening now and we'd like to talk with you. Sadia Raveendran, Surya Swamy, Sanjay Singh, PhD and I are here to HELP!!!! #GTC2026 #AIInfrastructure #AIInference #Energy #Datacenters #GPUs #Power #HammerheadAI #InferenceEconomy #inference The GTM Firm NVIDIA
1 Comment
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content