- Semiconductor BusinessHOME
- Products and Services of Macnica,Inc.
-
technical information
-
Events and Seminars
- Handling Manufacturer
- Support
- Inquiry
- Click here to purchase products
- Semiconductor business e-mail magazine registration
Narrow down by specifying conditions
現在2247件がヒットしています。check
In this series, we have shown how to handle IoT sensor data with NVIDIA NemoClaw, based on a simple demo application. Although some challenges remain, we hope you have understood that it is possible to utilize IoT sensor data with NemoClaw without making any changes to NemoClaw itself and while keeping the user's software development effort low.
We introduced two methods for detecting changes in sensor data: a pull method where NemoClaw periodically polls for changes, and a push method where an external program notifies NemoClaw in real time. Compared to the basic pull method, the push method requires slightly more software development effort on the user's side, but we hope you understand that it is feasible. We expect that support for push-type use cases will become easier to implement in future releases of NemoClaw.
Our focus was on demonstrating that NemoClaw 's AI agent can handle sensor data, and we haven't yet reached actual use cases such as predictive maintenance of manufacturing equipment. However, we hope you will consider the demo we presented as a starting point for exploring real-world use cases.
Future challenges
From here, we will introduce some points to keep in mind when utilizing IoT sensor data, which were not covered in this series.
LLM
For this demo, we decided to use only one NVIDIA DGX Spark™ chip, excluding the microcontroller board on the sensor side. Therefore, we adopted the relatively small qwen3.6-35b-a3b-nvfp4 model for use with the AI agent.
LLM performance is constantly improving, and the optimal model is likely to continue evolving. On the other hand, compatibility with AI agents also needs to be considered. Generally, larger models tend to yield better results, but in this test, even models larger than the one used and that fit within the DGX Spark did not necessarily perform better. If an edge-only system configuration isn't strictly required, another option is to serve a large-scale LLM on the data center side and share it among multiple NemoClaw instances.
REST API Server vs. MCP Server
In this series, we created a new application-specific REST API server as a means for NemoClaw 's AI agent to acquire data. By using FastAPI, implementation was possible with minimal user code, and by leveraging coding AI, the skill to instruct how to use the REST API could be created in a short time, so the amount of effort required for software development was not a major challenge.
On the other hand, there are cases where it is possible to handle things with an MCP server without creating an application-specific REST API server. NemoClaw supports MCP servers, but we did not use them in the demos in this series due to security constraints. We hope that in the future, know-how for using MCP servers with NemoClaw will be accumulated and the related ecosystem will be enhanced, making it easier to utilize them.
Security
The demonstrations in this series are built according to NemoClaw 's network policy, but no measures such as encrypting the distribution of sensor data have been taken. This is not expected to be a major problem during the proof-of-concept stage, but security measures will need to be considered for actual operation.
In this five-part series, we introduced methods for handling IoT sensor data using NVIDIA NemoClaw. This series covered everything from data acquisition to its utilization by AI agents, and pull-type and push-type data integration, all illustrated with demonstrations. We hope this verification will be helpful when considering IoT systems and new use cases utilizing NemoClaw.
Inquiry
