Nvidia touts Jetson Thor kit for real-time robot reasoning
(2025/08/26)
- Reference: 1756162027
- News link: https://www.theregister.co.uk/2025/08/25/nvidia_touts_jetson_thor_kit/
- Source link:
Nvidia has released a new brain for humanoid robots called Jetson Thor that promises more compute power and more memory than its predecessor.
The Jetson Thor series of modules sport a 2070 FP4 TFLOPS Blackwell GPU, enabling the delivery of 7.5x more AI compute, 3.1x more CPU performance, and 2x more memory than Jetson Orin, which debuted in 2023.
Operating within a 130-watt power envelope, the [1]Jetson Thor series includes the Jetson AGX Thor Developer Kit, Jetson T5000, and Jetson T4000.
[2]
"This performance leap will enable roboticists to process high-speed sensor data and perform visual reasoning at the edge — workflows that were previously too slow to run in dynamic real-world environments," said Chen Su, senior technical product marketing manager at Nvidia, in a [3]blog post .
[4]
[5]
These robot modules are designed to process sensor data and AI inference with low latency, in order to assure the smooth operation of robots in real-world deployments.
Su points to customers like [6]Agility Robotics and Boston Dynamics that are planning to make use of Jetson Thor hardware.
[7]
Agility Robotics' fifth-generation Digit robot incorporates prior Jetson kit and the next iteration is expected to utilize the Thor revision.
Peggy Johnson, CEO of Agility Robotics, said in a statement that the edge processing capabilities offered by Jetson Thor will improve Digit's real-time responsiveness and enable the implementation of a broader set of skills.
[8]According to the company , robot body control, object manipulation, and processing perception data all require serious computing power for real-time operation.
[9]
"Thor will allow us to run larger, more powerful, and more intelligent policies and reasoning models on our robots locally, in customer facilities and deployments," the firm said.
[10]New Yorkers will soon be able to yell 'I'm walkin here!' to Waymo robotaxis
[11]Baidu robocabs break even on one metric in low-fare China, company expects to cash in elsewhere
[12]Open the pod bay door, GPT-4o
[13]How to train your robot: Wear a tiny one in a baby carrier
Boston Dynamics meanwhile is planning to integrate Jetson Thor into its Atlas humanoid robot. As you can seen in this recently released video demonstration, Atlas still isn't as adept as humans at basic object manipulation.
[14]Youtube Video
Su observed that Jetson Thor is built with current generative reasoning models in mind, including Cosmos Reason, DeepSeek, Llama, Gemini, and Qwen models, as well as those specifically intended for robotics like Isaac GR00T N1.5. The goal is to run these models in real time on the network edge in order to minimize the cloud as a dependency.
If tech leaders kept their word, we'd be awash in humanoid robots. The fact that Tesla's touted Optimus model remains a laboratory project says something both about the lack of consequences for hype and the enduring technical challenges of mimicking biomechanics honed by millions of years of evolution while sipping battery power and not toppling onto toddlers.
Still, work on humanoid robots continues and there is progress, albeit slow. Fabrice Noreils, a technical leader for L3/L4 highway automated vehicles with Valeo, recently published a [15]survey [PDF] of humanoid robot projects.
His paper enumerates the various challenges that still need to be addressed to make humanoid robots viable. These include mechanical control, manual dexterity, spatial reasoning, safety, operation and maintenance, integration with existing warehouse management systems, the need for cloud services, communication between the robot and those it has to interact with, and cost.
"If we take the analogy with self-driving cars, many of these companies promise a general purpose humanoid robot for 2025 or 2026 but it is very likely, as for autonomous cars, that the promises will not be kept and we will have to wait 5 or 10 years to see this type of robot actually operational," wrote Noreils. "However, it is likely that these robots will work in secure perimeters for a few years, while regulation entities define security standards for humanoid robots."
The Nvidia Jetson AGX Thor developer kit starts $3,499. And Nvidia Jetson T5000 modules start at $2,999 but you have to buy 1,000 units. ®
Get our [16]Tech Resources
[1] https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/#tech-specs
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://blogs.nvidia.com/blog/jetson-thor-physical-ai-edge/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.agilityrobotics.com/content/agility-robotics-powering-the-future-of-robotics-with-nvidia-jetson-thor
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.agilityrobotics.com/content/agility-robotics-powering-the-future-of-robotics-with-nvidia-jetson-thor
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[10] https://www.theregister.com/2025/08/22/new_york_waymo_ratso_rizzo/
[11] https://www.theregister.com/2025/08/21/baidu_q2_2025/
[12] https://www.theregister.com/2025/08/20/gpt4o_pod_bay_door/
[13] https://www.theregister.com/2025/08/05/train_robots_baby_carriers/
[14] https://www.youtube.com/watch?v=HYwekersccY
[15] https://arxiv.org/pdf/2404.04249
[16] https://whitepapers.theregister.com/
The Jetson Thor series of modules sport a 2070 FP4 TFLOPS Blackwell GPU, enabling the delivery of 7.5x more AI compute, 3.1x more CPU performance, and 2x more memory than Jetson Orin, which debuted in 2023.
Operating within a 130-watt power envelope, the [1]Jetson Thor series includes the Jetson AGX Thor Developer Kit, Jetson T5000, and Jetson T4000.
[2]
"This performance leap will enable roboticists to process high-speed sensor data and perform visual reasoning at the edge — workflows that were previously too slow to run in dynamic real-world environments," said Chen Su, senior technical product marketing manager at Nvidia, in a [3]blog post .
[4]
[5]
These robot modules are designed to process sensor data and AI inference with low latency, in order to assure the smooth operation of robots in real-world deployments.
Su points to customers like [6]Agility Robotics and Boston Dynamics that are planning to make use of Jetson Thor hardware.
[7]
Agility Robotics' fifth-generation Digit robot incorporates prior Jetson kit and the next iteration is expected to utilize the Thor revision.
Peggy Johnson, CEO of Agility Robotics, said in a statement that the edge processing capabilities offered by Jetson Thor will improve Digit's real-time responsiveness and enable the implementation of a broader set of skills.
[8]According to the company , robot body control, object manipulation, and processing perception data all require serious computing power for real-time operation.
[9]
"Thor will allow us to run larger, more powerful, and more intelligent policies and reasoning models on our robots locally, in customer facilities and deployments," the firm said.
[10]New Yorkers will soon be able to yell 'I'm walkin here!' to Waymo robotaxis
[11]Baidu robocabs break even on one metric in low-fare China, company expects to cash in elsewhere
[12]Open the pod bay door, GPT-4o
[13]How to train your robot: Wear a tiny one in a baby carrier
Boston Dynamics meanwhile is planning to integrate Jetson Thor into its Atlas humanoid robot. As you can seen in this recently released video demonstration, Atlas still isn't as adept as humans at basic object manipulation.
[14]Youtube Video
Su observed that Jetson Thor is built with current generative reasoning models in mind, including Cosmos Reason, DeepSeek, Llama, Gemini, and Qwen models, as well as those specifically intended for robotics like Isaac GR00T N1.5. The goal is to run these models in real time on the network edge in order to minimize the cloud as a dependency.
If tech leaders kept their word, we'd be awash in humanoid robots. The fact that Tesla's touted Optimus model remains a laboratory project says something both about the lack of consequences for hype and the enduring technical challenges of mimicking biomechanics honed by millions of years of evolution while sipping battery power and not toppling onto toddlers.
Still, work on humanoid robots continues and there is progress, albeit slow. Fabrice Noreils, a technical leader for L3/L4 highway automated vehicles with Valeo, recently published a [15]survey [PDF] of humanoid robot projects.
His paper enumerates the various challenges that still need to be addressed to make humanoid robots viable. These include mechanical control, manual dexterity, spatial reasoning, safety, operation and maintenance, integration with existing warehouse management systems, the need for cloud services, communication between the robot and those it has to interact with, and cost.
"If we take the analogy with self-driving cars, many of these companies promise a general purpose humanoid robot for 2025 or 2026 but it is very likely, as for autonomous cars, that the promises will not be kept and we will have to wait 5 or 10 years to see this type of robot actually operational," wrote Noreils. "However, it is likely that these robots will work in secure perimeters for a few years, while regulation entities define security standards for humanoid robots."
The Nvidia Jetson AGX Thor developer kit starts $3,499. And Nvidia Jetson T5000 modules start at $2,999 but you have to buy 1,000 units. ®
Get our [16]Tech Resources
[1] https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/#tech-specs
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://blogs.nvidia.com/blog/jetson-thor-physical-ai-edge/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.agilityrobotics.com/content/agility-robotics-powering-the-future-of-robotics-with-nvidia-jetson-thor
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.agilityrobotics.com/content/agility-robotics-powering-the-future-of-robotics-with-nvidia-jetson-thor
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/edgeiot&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aK0xWTAeBIxAZGLNCQRUVQAAAEE&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[10] https://www.theregister.com/2025/08/22/new_york_waymo_ratso_rizzo/
[11] https://www.theregister.com/2025/08/21/baidu_q2_2025/
[12] https://www.theregister.com/2025/08/20/gpt4o_pod_bay_door/
[13] https://www.theregister.com/2025/08/05/train_robots_baby_carriers/
[14] https://www.youtube.com/watch?v=HYwekersccY
[15] https://arxiv.org/pdf/2404.04249
[16] https://whitepapers.theregister.com/