Lujie Yang's Recent LinkedIn Posts

Humanoid motion tracking performance is greatly determined by retargeting quality! Introducing 𝗢𝗺𝗻𝗶𝗥𝗲𝘁𝗮𝗿𝗴𝗲𝘁🎯, generating high-quality interaction-preserving data from human motions for learning complex humanoid skills with 𝗺𝗶𝗻𝗶𝗺𝗮𝗹 RL: - 5 rewards, - 4 DR terms, - Proprio. ONLY, - NO history/curriculum. Ready for agile, human-like 🤖? (Best with 🎧) 🔗 https://lnkd.in/e_8eMMtp 🎥 We are open-sourcing over 4 hours of high-quality, retargeted trajectories! https://lnkd.in/eWwKKBJR Huge shout out to the amazing team: Xiaoyu Huang, Zhen Wu, Angjoo Kanazawa, Pieter Abbeel, Carmelo (Carlo) Sferrazza, Karen Liu, Rocky Duan, Guanya Shi from Amazon FAR (Frontier AI & Robotics).

Can humanoids perform 𝗮𝗴𝗶𝗹𝗲, 𝗮𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀, 𝐥𝐨𝐧𝐠-𝗵𝗼𝗿𝗶𝘇𝗼𝗻 parkour based on what they see in the world? We present 𝗣𝗲𝗿𝗰𝗲𝗽𝘁𝗶𝘃𝗲 𝗛𝘂𝗺𝗮𝗻𝗼𝗶𝗱 𝗣𝗮𝗿𝗸𝗼𝘂𝗿 (𝗣𝗛𝗣): a modular framework that chains dynamic human skills using onboard depth perception. Goal: perception in → parkour actions out. Method: long-horizon trajectory synthesis via motion matching → RL motion tracking → large-scale depth-conditioned, multi-skill distillation. Try the policy yourself! https://lnkd.in/geiV95vJ Project website: https://lnkd.in/gVx-Q8Wb Huge shout out to the amazing team: Zhen Wu, Xiaoyu Huang, Yuanhang Zhang, Koushil Sreenath, Peter Chen, Pieter Abbeel, Rocky Duan, Angjoo Kanazawa, Carmelo (Carlo) Sferrazza, Guanya Shi, Karen Liu from Amazon FAR (Frontier AI & Robotics), UC Berkeley, Stanford, and CMU! This work was done during an internship at Amazon FAR.

I've long wondered if we can make a humanoid robot do a 𝘄𝗮𝗹𝗹𝗳𝗹𝗶𝗽 - and we just made it happen by leveraging 𝗢𝗺𝗻𝗶𝗥𝗲𝘁𝗮𝗿𝗴𝗲𝘁 with BeyondMimic tracking! This came after our original OmniRetarget experiments, with only minor tweaks to RL training: relaxing a termination threshold and removing one reward term. The policy achieved a 𝟱/𝟱 success rate in our real-world experiments, showing the strength of high-quality, interaction-preserving motion retargeting combined with minimal RL tracking. Here is the updated arXiv: https://lnkd.in/eNtWz_fz (In Sec. V. A) and website: https://lnkd.in/e_8eMMtp. Original post: https://lnkd.in/eNMZfQ88.