Sehoon Ha

We believe the best way to move forward is to bring others along. That's why we're launching our first @DexmateAI Research Grant Program! Selected research proposals will receive access to Vega U, our dual-arm manipulation platform, to explore and advance the frontiers of physical AI and embodied intelligence. Who's eligible: Full-time faculty at accredited universities or degree-granting research institutions with a US mailing address. 📅 Deadline: April 5, 11:59 PM PST 🔗 Apply: dexmate.ai/research-grant Not a PI? Pass this along to your lab lead — and tag anyone who should apply! #Dexmate #VegaU #Robotics #PhysicalAI #EmbodiedAI #ResearchGrant

Introducing EgoVerse: an ecosystem for robot learning from egocentric human data. Built and tested by 4 research labs + 3 industry partners, EgoVerse enables both science and scaling 1300+ hrs, 240 scenes, 2000+ tasks, and growing Dataset design, findings, and ecosystem 🧵

Decisions for @CVPR 2026 are out—congratulations to all authors. I’m excited to share a community step forward: the new CVPR Findings Track. Area Chairs recommended 1717 papers for potential inclusion, creating a principled pathway to recognize and share valuable work that may not be the best fit for the main program—while still enabling authors to publish and present through integrated Findings poster sessions. As our field scales, we need not only better models—but better community infrastructure. This effort is led collectively by the Findings organizing team—Bryan Plummer, Kevin Shih, @anand_bhattad, @jccaicedo, @Grigoris_c, @BoqingGo, @liuziwei7, and me. Huge thanks to the CVPR General Chairs, Program Chairs, and especially the Area Chairs for supporting this step forward. Looking forward to seeing many of you at CVPR 2026—across the main program, Findings, and workshops.

A new AI job coach from @GeorgiaTech aims to help neurodivergent job seekers. @jennifergkim discusses the tool's value for job applicants + employers. Kim also talks about her personal motivations with the work, having a brother with autism. Video 📽️: 11alive.com/video/news/ai-…

Georgia Tech’s School of Interactive Computing is launching a major search for a new faculty member in Computer Graphics this year. Despite a challenging job market, we are committed to recruiting an exceptional colleague who will help shape the future of our graphics community. We are especially interested in candidates who • have a strong presence in the Computer Graphics / SIGGRAPH community, and • bring research directions that complement—not overlap with—our current faculty (Sehoon Ha, Bo Zhu, Greg Turk). If you know outstanding scholars, we would be grateful for your recommendations. We warmly encourage applicants from all backgrounds and institutions. 📌 Official posting: ic.gatech.edu/faculty-hiring 📩 Contact: sehoonha@gatech.edu / bo.zhu@gatech.edu / turk@cc.gatech.edu

See Spot perform dynamic whole-body manipulation. Using a combination of reinforcement learning (RL) and sampling-based control, the robot is able to autonomously drag, roll, and stack tires weighing 15 kg (33 lb), well above its maximum arm lift capacity. Learn more about coordinating locomotion and manipulation processes: rai-inst.com/resources/blog…

We taught Spot to stack 15kg car tires autonomously. It uses its whole-body and shows some dynamic manipulation!

For more details on our method and results, please refer to our paper that will show up on our project website: 🔗 hyunyoungjung.github.io/projects/ppf/p…

✅ With RL + MAR, our controller achieves robust locomotion across diverse indoor & outdoor experiments, demonstrating a forward speed of 1.5 m/s and robust locomotion across diverse terrains, including slippery, sloped, uneven, and sandy terrains.

What if we can take the periodic style of model-based controllers and the robustness of learning-based policies? In our #RAL paper, PPF: Pre-training and Preservative Fine-tuning of Humanoid Locomotion via Model-Assumption-based Regularization, we address this challenge. 🔑 Key idea: Imitate the model-based controller and then Fine-tune with RL, while preserving motion style via Model-Assumption-based Regularization (MAR) to avoid forgetting. Our controller is verified through comprehensive simulation tests and hardware experiments on a full-size humanoid robot, Digit, demonstrating a forward speed of 1.5 m/s and robust locomotion across diverse terrains, including slippery, sloped, uneven, and sandy terrains. 🙋Authors: Hyunyoung Jung* @hyunyoungjung, Zhaoyuan Gu* @gu_zy14, Ye Zhao @GT_LIDAR, Hae-Won Park, and Sehoon Ha @sehoonha 🌐 Website: hyunyoungjung.github.io/projects/ppf/p… 📄 arXiv: arxiv.org/abs/2508.134443

📢 The Proceedings of the 2025 Conference on Robot Learning (CoRL 2025) are now available through PMLR! Discover the latest advances in robot learning research presented at this year’s conference. 🔗 proceedings.mlr.press/v305 #CoRL2025 #RobotLearning #PMLR #MachineLearning #Robotics

We've arrived in the Top 5! The 2026 @usnews Best Colleges rankings place Georgia Tech's CS program at No. 5 in the U.S. (up from No. 7 last year). #ChangeTheGame is part of a culture that's helped us reach this milestone. Go #GTComputing! Read more: cc.gatech.edu/news/ai-and-cy…

📢 The CoRL 2025 accepted papers are now live on OpenReview! Check out the list of exciting contributions in robot learning: openreview.net/group?id=robot… #CoRL2025 #RobotLearning

🚨 Paper Alert 🚨 ➡️Paper Title: EMMA: Scaling Mobile Manipulation via Egocentric Human Data 🌟Few pointers from the paper 🎯Scaling mobile manipulation imitation learning is bottlenecked by expensive mobile robot teleoperation. 🎯Authors of this paper presented “Egocentric Mobile MAnipulation (EMMA)”, an end-to-end framework training mobile manipulation policies from human mobile manipulation data with static robot data, sidestepping mobile teleoperation. 🎯To accomplish this, they co-train human full-body motion data with static robot data. 🎯In their experiments across three real-world tasks, EMMA demonstrates comparable performance to baselines trained on teleoperated mobile robot data (Mobile ALOHA), achieving higher or equivalent task performance in full task success. 🎯They found that EMMA is able to generalize to new spatial configurations and scenes, and they observe positive performance scaling as they increase the hours of human data, opening new avenues for scalable robotic learning in real-world environments. 🏢Organization: @GeorgiaTech 🧙Paper Authors: Lawrence Y. Zhu, Pranav Kuppili, @ryan_punamiya , Patcharapong Aphiwetsa, Dhruv Patel, @simar_kareer , @sehoonha , @danfei_xu 📝 Read the Full Paper here: arxiv.org/abs/2509.04443 🗂️ Project Page: ego-moma.github.io 🎥 Be sure to watch the attached Demo Video - Sound on 🔊🔊 Find this Valuable 💎 ? ♻️QT and teach your network something new Follow me 👣, @NaveenManwani17 , for the latest updates on Tech and AI-related news, insightful research papers, and exciting announcements.

As a result, SDAX allows an agent to learn challenging motor skills without reward shaping, demonstrations, or curricula. We showcase hardware results for crawl, leap, and climb, and simulation results for super-agile wall-jumping. For more details, please visit our website: seungeunrho.github.io/projects/SDAX/

“Skill collapse—but in a positive way.” One challenge in skill discovery is selecting the right skill from a distribution once training ends. With SDAX, task rewards act like a magnet, aligning diverse skills toward solving the task. We call this positive skill collapse—thanks to it, we could easily pick the right learned skills.

Can robots learn to jump, crawl, and climb—without reward engineering, curriculum learning, or reference data? In our #CoRL2025 paper SDAX, Skill Discovery as eXploration, we show how unsupervised RL can learn these challenging motor skills without human priors —and transfer them to real robots. 📷 Unsupervised Skill Discovery as Exploration for Learning Agile Locomotion Seungeun Rho* @ross_rho , Kartik Garg* @_pacificax, Morgan Byrd, Sehoon Ha @sehoonha seungeunrho.github.io/projects/SDAX/