Spirit AI’s latest embodied intelligence model, Spirit v1.5, just smashed the global RoboChallenge with a top score of 66.09 and a 50.33% success rate—making it the only robot brain to clear the 50% mark! 🤯
What’s RoboChallenge? Think of it as the ultimate “exam” for real-world robots: a live testbed with 30 everyday tasks—like placing objects, recognizing targets and even using tools 🛠️. Spirit v1.5 aced these challenges, beating out top contenders from the U.S.
Even cooler: Spirit AI has open-sourced the model and all the resources, so developers everywhere can dive in and build on their breakthrough 🔓✨
Based in Hangzhou—home to DeepSeek and Unitree Robotics—Spirit AI is all about making robots that can see, think and act in the physical world. Last June, they rolled out Moz1, a humanoid bot aimed at logistics and industrial use 🤖📦
“This result shows strong all-around robot skills and real-world readiness,” says Qiu Jiefan from Zhejiang University of Technology. It’s not mass deployment…yet—but it’s a huge step forward! ✔️
Han Fengtao, Spirit AI’s CEO, explains that their Vision-Language-Action (VLA) architecture merges perception, reasoning and action into one system—cutting out error-prone handoffs and boosting performance.
Keep an eye out: he predicts a wave of smarter service robots in the next 2–3 years, ready to help in warehouses, hospitals and beyond. The future just got closer! 🚀
Reference(s):
Chinese AI startup tops global embodied intelligence benchmark
cgtn.com
