Intelligent surface classication for radar based mobile robotics
Our radar solutions supports Edge AI applications from a single sensor that accurately differentiates several surfaces in real world applications.
Application overview
Traditional sensing technologies like PIR, UWB, and ToF typically only provide basic presence information. Our radar sensing solutions leverage Edge AI to enable advanced sensing capabilities like surface classification. Allowing mobile robots to identify the surface beneath them such as grass, concrete, or other terrain and respond accordingly in real time.
Powered by TI’s low power, cost optimized IWRL6432 radar, engineers can easily enable advanced surface classification features to the next generation of mobile robotics.
Starting evaluation
Data collection
Capture accurate radar data with the IWRL6432 devices, enabling full AI processing done completely on the edge through the integrated M4F MCU and HWA.
Data quality assessment
TI radar sensors support a wide range of output formats, including range profiles, range-doppler heatmaps, point clouds, and more. For surface classification, the model leverages range profile data to achieve high accuracy performance.
If you would like to learn more about the different types of Radar data and output formats, you can get started today with our Radar Academy.
Customers can get started immediately using the sample dataset available in CCStudio™ Edge AI Studio in which you can can find tools that allows users to upload unfiltered range profile data and experiment with different range windows to optimize performance, as detailed in the surface classification user guide.
Build and train your model
The radar extension in CCStudio™ Edge AI Studio is built around two key components:
- A front end GUI, available on a local system or cloud server
- A back-end PyTorch engine that handles API calls and model execution.
Beginners can start using our CCStudio™ Edge AI Studio to explore the radar surface classification example project. This guided experience includes a radar pre-loaded dataset evaluation, model definition, compilation, and other key steps in the workflow.
More advanced users who are comfortable with radar sensing and machine learning can interact directly with the back-end engine.
Training, compilation, and evaluation are fully customizable, allowing users to replace individual steps with their own methods as needed.
This flexible architecture ensures the tool delivers value to users at every level from those new to radar and machine learning to experienced developers seeking automation and integration guidance for deploying models into our radar firmware projects.
Follow our step-by-step video guide on how to get started with TI’s surface classification example model and how to collect, train and deploy your data quickly.
Deploying your model
Once trained and validated, CCStudio™ Edge AI Studio provides an end-to-end workflow for deploying trained models directly to our radar devices without any other systems needed.
All the hardware, software and resources you’ll need to get started
Hardware
IWRL6432
Single-chip low-power 57-GHz to 64-GHz industrial mmWave radar sensor
Software & development tools
Radar Toolbox for Edge AI
Learn more about how TI’s 60GHz and 77GHz industrial mmWave sensors can measure distance and relative velocities of people or objects.
CCStudio™ Edge AI Studio
Edge AI studio contains tools for training, compiling and deploying a model to TI Edge AI processors.
Supporting resources
Learn more about TI’s different radar data outputs are used and learn more about radar as a technology.