Zhi Su
@ZhiSu22
Undergrad @Tsinghua_IIIS | 🤖 Robot Learning | PhD '26 apps
1️⃣/7️⃣🤖⚽ Toward Real-World Cooperative and Competitive Soccer with Quadrupedal Robot Teams Most learning-based robot-soccer work stays in simulation or tests 1v1. We field real-world games with both cooperation and competition—plus robot-human matches!
I'm observing a mini Moravec's paradox within robotics: gymnastics that are difficult for humans are much easier for robots than "unsexy" tasks like cooking, cleaning, and assembling. It leads to a cognitive dissonance for people outside the field, "so, robots can parkour &…
Today, We’re launching Genesis AI — a global physical AI lab and full-stack robotics company — to build generalist robots and unlock unlimited physical labor. We’re backed by $105M in seed funding from @EclipseVentures, @khoslaventures, @Bpifrance, HSG, and visionaries…
With high-fidelity simulation and ray-tracing rendering, we can minimize the sim-to-real gap and achieve zero-shot sim-to-real transfer! We hope this is a critical step for scaling humanoid-specific data that is scarce atm.
🚀 Introducing LeVERB, the first 𝗹𝗮𝘁𝗲𝗻𝘁 𝘄𝗵𝗼𝗹𝗲-𝗯𝗼𝗱𝘆 𝗵𝘂𝗺𝗮𝗻𝗼𝗶𝗱 𝗩𝗟𝗔 (upper- & lower-body), trained on sim data and zero-shot deployed. Addressing interactive tasks: navigation, sitting, locomotion with verbal instruction. 🧵 ember-lab-berkeley.github.io/LeVERB-Website/
Glad to share my work during my internship at CMU! Let’s push agility, adaptivty and safety further forward! Kudus to my mentors @TairanHe99 @ChongZitaZhang and advisor @GuanyaShi
It might be a bit late but here's our work in L4DC 2025 on bridging agility, safety, and adaptivity. Compared to prior work (ABS) that safeguards agile locomotion, this one further explicitly estimates the change in dynamics and does adaptive shielding. adaptive-safe-locomotion.github.io
3/3👇 ❌ Async RL introduces data staleness & Classic PPO collapses. 🧵 Solution? 1️⃣ System-level staleness control for max-allowed staleness 2️⃣ Decoupled PPO objective for off-policy data ✅ AReaL allows highly stale data w.o. perf. drop 📊 Ablation w. AReaL&PPO on math RL👇
Start an internship with @kevin_zakka at Boston Dynamics and work on the E-Atlas RL control. I’m excited about the opportunity to conduct research with the most advanced robot and talented team here.
Try out @Astribot_Inc's teleop system at #ICRA2025 — it's incredibly smooth, with almost no latency, and remarkably easy to use! I can’t imagine how much low-level control optimization went into achieving this.
🧠+🐶 = Terrain Navigation Introducing SARO: a Space-Aware Robot System that combines Vision-Language Models and RL-based control for robust quadruped navigation in 3D terrains 🌄 Catch us at Thursday 15:15pm at Room 309 👉 saro-vlm.github.io
🚨 New work at #ICRA2025! Robust Robot Walker 🐾 We enable quadruped robots to pass tiny traps (bars, pits, poles) using only proprioception – no cameras, no depth! Catch us at Thursday 16:55pm in Room 305! 🔗 robust-robot-walker.github.io
Congrats, dude! Impressive work!
1️⃣/7️⃣🤖⚽ Toward Real-World Cooperative and Competitive Soccer with Quadrupedal Robot Teams Most learning-based robot-soccer work stays in simulation or tests 1v1. We field real-world games with both cooperation and competition—plus robot-human matches!
Very amazing cooperative agents!
1️⃣/7️⃣🤖⚽ Toward Real-World Cooperative and Competitive Soccer with Quadrupedal Robot Teams Most learning-based robot-soccer work stays in simulation or tests 1v1. We field real-world games with both cooperation and competition—plus robot-human matches!
Great to meet @DrJimFan at #ICRA2025 — after only seeing your YouTube talks till now.
Multi-agent RL enables quadruped soccer games and human-robot cooperation in the real world!
1️⃣/7️⃣🤖⚽ Toward Real-World Cooperative and Competitive Soccer with Quadrupedal Robot Teams Most learning-based robot-soccer work stays in simulation or tests 1v1. We field real-world games with both cooperation and competition—plus robot-human matches!
Robot Soccer!🤖⚽ Thrilled to share our first step on multiple quadrupedal robot soccer in the real world! All behaviors, defending, attacking, counterattacks, team coordination, are learned entirely through self-play! Looking forward to full-scale robot soccer in real!
1️⃣/7️⃣🤖⚽ Toward Real-World Cooperative and Competitive Soccer with Quadrupedal Robot Teams Most learning-based robot-soccer work stays in simulation or tests 1v1. We field real-world games with both cooperation and competition—plus robot-human matches!
Very cool drone shot
1️⃣/7️⃣🤖⚽ Toward Real-World Cooperative and Competitive Soccer with Quadrupedal Robot Teams Most learning-based robot-soccer work stays in simulation or tests 1v1. We field real-world games with both cooperation and competition—plus robot-human matches!
Check out Zhi's thread to see how quadrupedal robots can play soccer as a team—it’s truly amazing to see the experiments in person! Zhi is also currently seeking a Ph.D. position for Fall 2026. He’s at ICRA now, so if you’re interested in his work, feel free to reach out to him!
1️⃣/7️⃣🤖⚽ Toward Real-World Cooperative and Competitive Soccer with Quadrupedal Robot Teams Most learning-based robot-soccer work stays in simulation or tests 1v1. We field real-world games with both cooperation and competition—plus robot-human matches!
Human & robot dog cooperative soccer game with multi-agent RL! Check out the video ;)
7️⃣/7️⃣ 📄 Paper: arxiv.org/abs/2505.13834 🎬 Full video: youtu.be/7gq7N16jKgI 🙏 Big thanks to my co-author Yuman Gao and to our mentors @ZhongyuLi4, @jxwuyi, and @KoushilSreenath for their invaluable guidance and support! #Robotics #Quadruped #RobotSoccer