NVIDIA Jetson Thor accelerates embodied intelligent universal processes

Kuafu ", as a representative full-size humanoid robot of Leju, has more than 40 high degree of freedom full body biomimetic structures, providing greater flexibility and adaptability for the execution of complex tasks. However, this high degree of freedom design also brings higher data computing requirements, and the system needs to coordinate the coordinated motion of all body joints in real time to accurately complete various operational tasks while maintaining overall balance.

NVIDIA Jetson Thor， With the 2070 FP4 TFLOPS AI computing power, lightweight, flexible deployment, and multimodal perception support, it assists "Kuafu" in presenting excellent computing capabilities on ACT, DP, VLA, Pi0 and other models, contributing to smoother and smoother operational performance in industrial scenarios such as logistics, intelligent manufacturing, 3C electronics, automotive assembly, and daily chemical production training.

In logistics scenarios, the shape, size, and material of items are not fixed, and robots need to quickly and accurately identify labels and grasp targets. They also need to maintain stable performance and consistent task execution during long-term continuous operations.

To avoid the degradation of accuracy and speed over time, NVIDIA Jetson Thor has increased its bandwidth by 35%, effectively helping "Kuafu" to handle multi-modal large traffic stably for a long time, achieving continuous and stable operation of intelligent optimization algorithms, iterative learning control and other algorithm models.

This also means that "Kuafu" can use multimodal perception in long-term sorting execution, learn and correct errors, maintain and improve accuracy in real environments, and achieve stable and durable performance breakthroughs.