Bring Ease to the World with Physical AI.
- Tweets 13
- Following 1
- Followers 20
- Likes 1
Congrats to Prof. @ChaoYuTHU on being named a 2025 TR 35 Asia Pacific.🏆🎉
Her MAPPO work and large-scale RL infrastructure for embodied intelligence are already shaping how we train robots that act in the real world.🤖🧠
#MIT #TR35 #InnovatorsUnder35 #Rlinf
Honored to be named a 2025 TR35 Asia Pacific 🏆
Thank you to everyone who has supported this journey.
🚀Onward with better AI models & RL systems for embodied intelligence.
#MIT #TR35
59
How can robots clear a cluttered tabletop? 🤖
Our Physical AI system enables the robot to recognize each object, decides where it belongs, and completes a long-horizon sequence of actions—including coordinated dual-arm handovers.
Thread 👇
1
2
140
Clearing a cluttered table is more than a pick-and-place task.
The task demonstrates three core capabilities:
1️⃣Semantic understanding of everyday objects
2️⃣Long-horizon planning and execution
3️⃣Coordinated dual-arm manipulation and handovers
1
86
Higher task success and faster execution. 🚀
Our Physical AI system completes a multi-stage unboxing task with tool use & high-precision bimanual manipulation.
Powered by #STEAM , it learns frame-level task progress from mixed-quality real-world experience.
2
4
458
From learning trajectories → learning progress.
@ChaoYuTHU proposed STEAM, a label-free advantage model that reads progress, stalls, failures, and recoveries frame‑by‑frame — learned from nothing but the temporal order of expert demos.
1
1
365
Real-world gains 🚀
On four real-robot tasks, paired with CFGRL, STEAM lifts success rate by +59%, +54.3%, +23%, and +16.2% — towel folding, chip checkout, cola restocking, and pick-and-place.
🌐 Project: rlinf.github.io/steam/
66
We see a possible paradigm shift:
End-to-End VLA → Harness VLA
@ChaoYuTHU open-sourced Harness VLA — adding a Harness Layer to organize VLA models without changing their weights.
LIBERO-Pro:
Harness VLA 82.4% 🚀
vs.
Pi_RLinf 50%
NVIDIA Cap-X 18.2%
Berkeley RATS 43.8%
1
1
91
Harness VLA improves reliability because exploration teaches the planner how to use a fixed primitive set: where to ground entities, when to invoke VLA_ACT, how to re-stage failed contacts, and which failures should not be repeated.
1
56
How can robots help people at the checkout? 🛒
Here's another Striding AI's demo as we work toward a future where humans and robots collaborate seamlessly.
#PhysicalAI #Robotics #AI #EmbodiedAI
77
👀 A sneak peek at Striding AI’s Physical AI system performing autonomous shelf restocking in a simulated retail environment.
#PhysicalAI #Robotics #AI #StridingAI #EmbodiedAI #MachineLearning
1,420
Introducing Striding AI. 💫
Our mission is simple:
Bring Ease to the World with #PhysicalAI.
What will the future look like?
We envision robots becoming trusted teammates—helping people work safer, smarter, and more efficiently. 💙
Explore more: striding.ai
1
111