An Introduction to Building Humanoid Robots | NVIDIA GTC 2025

[Alan Timm]

https://www.youtube.com/watch?v=Oyon1QDpU6g

Useful humanoid (and non-humanoid) robots are happening now, and almost all of them are using an nvidia development training and inference framework.  This is exactly what Alfie's been waiting for.

In addition to insights about designing, building, and training humanoid robots, Nvidia dropped this nugget about their next-gen Jetson Thor AGX platform, which is going to be released in a month or two.

Jetson Thor AGX

• Available in June 2025
• 2560 CUDA cores, 96 Tensor cores (+25% from Orin AGX)
• 7.8 FP32 TFLOPS (47% faster than Jetson Orin AGX at 5.32 FP32 TFLOPS)
• 2000 FP4 TOPS
• 1000 FP8 TOPS (Orin AGX is 275 INT8 TOPS; Blackwell has same INT8/FP8 performance)
• 14 ARMv9 cores at 2.6x performance of Orin cores (Orin has 12 cores)
• 128GB of RAM (Orin AGX is 64GB)
• 273GB/s RAM bandwidth (33% faster than Orin AGX at 204.8GB/s)
• 120W max power (double Orin AGX at 60W)
• 4x 25GbE
• 1x 5GbE (at least present on devkit)
• 12 lanes PCIe Gen5 (32GT/s per lane).
• 100mm x 87mm (same as existing AGX)
• All I/O interfaces for devkit "on one side of board"
• Integrated 1TB NVMe storage on devkit

[Donny Lu]

Thanks for sharing this! Planning to work on a few bipedal robotics project here at Caltech robotics team! Love to learn that Nvidia provides supports for them!

--
You received this message because you are subscribed to the Google Groups "RSSC-List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rssc-list+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/rssc-list/b50ba14a-847f-48ad-92c1-4f52bd4db9fen%40googlegroups.com.