Nov - Week 3- Robotix Pulse

𝗩𝗣 (Vice President) Career Path Study: 𝗣𝗮𝗿𝘁 𝟮: What Did Tech VPs Study & what School did they graduate from? Last week in Post #1, we kicked off a series where we analyze ~100 LinkedIn profiles of VPs from top tech companies: 𝘈𝘱𝘱𝘭𝘦, 𝘕𝘝𝘐𝘋𝘐𝘈, 𝘎𝘰𝘰𝘨𝘭𝘦, 𝘈𝘮𝘢𝘻𝘰𝘯, 𝘔𝘪𝘤𝘳𝘰𝘴𝘰𝘧𝘵, 𝘔𝘦𝘵𝘢, 𝘛𝘦𝘴𝘭𝘢, 𝘐𝘯𝘵𝘦𝘭, 𝘢𝘯𝘥 𝘓𝘪𝘯𝘬𝘦𝘥𝘐𝘯. Our 𝗴𝗼𝗮𝗹 is simple: Understand who these leaders are, their background, education, and the paths they took. In Part 1 [https://lnkd.in/ePBjhaCT], we shared two surprising insights: - 86.3% of VPs do not have an MBA - Their highest degrees lean heavily toward a graduate degree (master’s & PhD) 🎓 𝗣𝗮𝗿𝘁 𝟮: 𝗪𝗵𝗮𝘁 𝗗𝗶𝗱 𝗧𝗵𝗲𝘀𝗲 𝗩𝗣𝘀 𝗦𝘁𝘂𝗱𝘆? Based on our dataset, here are the top majors for tech VPs (grouped & normalized): - CS/CE (Computer Science / Computer Engineering) – largest share by far - EE (Electrical Engineering) - ME (Mechanical Engineering) 🏫 𝗪𝗵𝗲𝗿𝗲 𝗗𝗶𝗱 𝗧𝗵𝗲𝘆 𝗦𝘁𝘂𝗱𝘆? While Stanford and UC Berkeley are top U.S. universities they got their highest degree from, there is 𝗻𝗼 single “must-attend” university. Yes, top schools show up, but the majority come from a wide range of universities, including many you wouldn’t expect. 📌 From an 𝗲𝗱𝘂𝗰𝗮𝘁𝗶𝗼𝗻 perspective, our dataset suggests that MBAs are relatively uncommon among these VPs, and school names vary widely. The most frequent pattern we see is VPs holding graduate-level technical degrees, particularly in CS/CE, though this may not represent the entire industry ---> Follow along: more insights coming in this series. At the end of this series, we’ll release the full dataset ------------------------------------------------------------------------------------- For readers interested in upskilling in AI, Udacity currently has a 55% discount on its online 𝗠𝗮𝘀𝘁𝗲𝗿’𝘀 𝗶𝗻 𝗔𝗜 using code BLACKFRIDAY55 [sponsored & affiliate link]: https://lnkd.in/eTMFKN72 ----------------------------------------------------------------------------------------

𝗜𝗻-𝗡-𝗢𝗻: Teaching #robots from real first-person human videos Robots usually learn from small, clean datasets collected in labs. But real life is messy. Humans have thousands of hours of first-person videos showing how we grab objects, cook, clean, open drawers, and do everyday tasks. 𝗧𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺? This data is heterogeneous (all different and inconsistent), different cameras, different tasks, different environments. Most robot systems don’t know how to learn from it properly. 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻? Researchers at UC San Diego provide a recipe for collecting and using egocentric data (first-person human video showing what the person sees and does). They combine two kinds of human data: In-the-wild videos — everyday first-person footage On-task videos — videos that match robot manipulation tasks These form a new dataset called PHSD. They then train a model called Human0, which learns from video, hand motion, and language instructions. 𝗛𝗼𝘄 𝗶𝘁 𝘄𝗼𝗿𝗸𝘀 (super simplified): The robot “watches” first-person human videos. It learns how hands move and how objects respond. With a small amount of robot data, it can perform many tasks by itself. 𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀? Robots can learn more skills without needing tons of robot demonstrations. They understand human actions better. They become more general and robust 𝗞𝗲𝘆 𝗿𝗲𝘀𝘂𝗹𝘁𝘀 Human0 can follow language instructions even when trained mostly on human video. It learns new tasks with very little robot data (few-shot). Mixing in-the-wild + on-task data makes the robot more reliable. This is a step toward robots that learn the way humans do, by watching real people in real situations. 📖 Paper: https://lnkd.in/erJqxvyh 🌐Project Site: https://lnkd.in/e5Nh8SHU 👥 Authors: Xiongyi Cai, Ri-Zhao Qiu, Geng Chen, Lai Wei, Isabella Liu, Tianshu Huang, Xuxin Cheng, Xiaolong Wang --------------------------------------------------------------------------------- Udacity’s Fully online 𝗠𝗮𝘀𝘁𝗲𝗿’𝘀 𝗶𝗻 𝗔𝗜*: project-based, and built for real industry skills*--> https://lnkd.in/eTMFKN72 55% OFF -> Code “BLACKFRIDAY55” ---------------------------------------------------------------------------------- Thank you Ri-Zhao Qiu for the permission to use the video.

𝙇𝙤𝙘𝙤𝙏𝙤𝙪𝙘𝙝: teaching quadruped robots to “feel” while carrying objects Quadruped robots are getting better at walking and climbing stairs. But there’s one big challenge we don’t talk about enough: 𝘔𝘰𝘴𝘵 𝘭𝘦𝘨𝘨𝘦𝘥 𝘳𝘰𝘣𝘰𝘵𝘴 𝘤𝘢𝘯’𝘵 𝘧𝘦𝘦𝘭 𝘸𝘩𝘢𝘵 𝘵𝘩𝘦𝘺’𝘳𝘦 𝘤𝘢𝘳𝘳𝘺𝘪𝘯𝘨. If you put a box, or a fragile object it has no idea how the load is shifting. This makes transport tasks unsafe and unstable. LocoTouch developed by researchers at Carnegie Mellon University, University of Washington, and Google DeepMind tries to solve this problem. 𝗧𝗵𝗲 𝗰𝗼𝗿𝗲 𝗶𝗱𝗲𝗮 (simplified) LocoTouch adds tactile sensing to a quadruped robot so it can feel the forces from the object it’s carrying. And then they train a reinforcement learning (RL) policy that learns how to move dynamically while reacting to those forces. The combination is powerful: The robot senses pressure and force changes on its back RL learns how to adjust its gait, balance, and speed The robot can safely move even when the load shifts suddenly Basically: LocoTouch = “touch + RL locomotion” → safe dynamic transport. With LocoTouch, the robot can transfer zero-shot (performing new tasks without specific training) from simulation to the real world and safely transport many everyday objects with good robustness and agility. ------------------------------------------------------------------------------- 📰Paper: https://lnkd.in/gyS4nNJT 🌐Project Page: https://lnkd.in/gzze498A 🧑‍💻Code: https://lnkd.in/gsSG54Nx 👥 Authors: Changyi Lin, Yuxin Ray Song, Boda Huo, Mingyang Yu, Yikai Wang, Shiqi Liu, Yuxiang Yang, Wenhao Yu, Tingnan Zhang, Jie Tan, Yiyue Luo, Ding Zhao ----------------------------------------------------------------------------------- Advance your career with an accredited 𝗠𝗮𝘀𝘁𝗲𝗿’𝘀 𝗶𝗻 𝗔𝗜*: flexible, project-based, and built for real industry skills*: https://lnkd.in/eTMFKN72 40% off 𝘸𝘪𝘵𝘩 𝘤𝘰𝘥𝘦 “SINA40” *Thank you Udacity for sponsoring this post *This post contains an optional affiliate link that supports my work at no extra cost to you. -----------------------------------------------------------------------------------

Share your total compensation details (base salary, bonuses, RSUs) to help build a transparent resource for robotics professionals. Submissions are anonymous and will help others understand industry standards. By submitting, you confirm that your information doesn’t violate any confidentiality agreements or policies. Contributors get free access to the final database—thank you for your support! (you can email me separately after you submitted this form: robotixwithsina@gmail.com)