How to train and update your models faster?

Introducing Teraki’s DevCenter

DevCenter is Teraki’s free and one-stop-shop portal for customers to test Teraki’s technology with their own sensor data. Next to testing the Teraki technology, DevCenter is also used by customers in their production environments as a tool to train and add new sensors or signals to their data collection campaigns. In minutes, customers can easily train any new sensor-signal that they want to collect from their fleets. It is an easy and flexible way to reduce high amounts of raw sensor data of any type and to get detailed insight into the main KPI’s: reduction efficiency, accuracy and latency.

Teraki DevCenter Why?

DevCenter was created so that Teraki’s customers can easily and quickly get informed about the currently supported data reduction capabilities. Currently available for telematics data. Soon to be complemented with camera and 3D point cloud (lidar/radar/ToF) data.

Customers can “test drive” DevCenter by bringing their own, raw sensor data for processing in DevCenter. This way they can quickly assess the impact on their use cases whether it is focused on reducing data transmission or storage or improving a perception stack to solve a particular problem. The ability to extract >10X more data from a car per day enables our customers to train data models far quicker. The ability to extract 10X higher frequency (=better resolution) from a car enables our customers to train data models far more accurate. The resulting models can be downloaded and used for pre-processing the data directly in the car, prior for the data to be fed into a perception stack. Applying this pre-processing speeds up the inference time of the perception stack .

Furthermore, DevCenter is also used by our customers in production environments, i.e. not for testing Teraki technology but for managing live data flows from large fleets of cars. This is particularly relevant if customers want to collect new signals from a car; or when customers want to retrain their models. In such cases DevCenter automatizes this process in a very flexible and efficient manner.

In addition, customers get a sneak preview on the upcoming features that are being researched and developed in video and 3D point cloud areas.

What it Offers?

Currently, DevCenter supports data reduction for all kinds of telematics (1D) sensor data. The user can upload data files with sensor information gathered from sensors in the field e.g. speed, acceleration, magnetometer, gyroscope, etc. Once the raw sensor data is uploaded, the user can specify the signal types and trigger the training of Teraki’s data reduction model.

Graph Mileage city and highway

Figure 1 User Interface of the DevCenter.

After the training is complete, the user can visualize the original and the reconstructed signals, in addition to the key KPIs such as Maximum Deviation and data reduction rate. Customers can easily zoom in (and out) to see a more granular comparison between original and reconstructed signal.

In addition to that, DevCenter includes the documentation of the latest SDK Release.

It also includes a demo page. Currently the demo page shows a video with Teraki’s capabilities in reducing video (2D) data.

Graph Mileage city and highway

Testing Teraki’s “RoI” pre-processing for video.

Example Use Case: Predictive Maintenance

Generally, for a predictive maintenance use case a customer has to upload terabytes of data (think of speed, acceleration, temperature, motor speed, breaking, and other important sensor signals) to the cloud so that it can be processed. From there customers can start to predict when and what kind of maintenance would be required to a particular car (part).

With the help of DevCenter, the customer would bring these signals and upload them to it and can directly quantify how much data reduction would be achieved while maintaining the error rate within a pre-selected and acceptable limit. At the end of the training and with the help of the export tool in DevCenter they can easily extract the reconstructed signals and feed it into their predictive maintenance model to see whether there is any degradation in performance.

If the user is happy with all that, they can read through the documentation and request access to the SDK, so that a one-off and easy implementation can happen on the embedded device.

Potential Use Case: Crash Detection

When a road accident occurs, there might be a delay in sending the signals to police, ambulance and insurance companies for damage recoveries. Insurance companies can work more efficient if they have data driven crash reports available early in the claim management process. This will enable savings and efficiency in the otherwise time-consuming, manual process of claim management. DevCenter could act like a gateway where the insurance companies can detect any crashes happening and simplify and streamline the claim management process.

Example Use Case: Data Storage Management

The vast amount of data that is being generated by sensors in cars severely increases the challenges how to efficiently process this data. Data storage is costly at times and Teraki’s DevCenter helps in reducing the data by implementing Teraki’s trained models without degradation in the quality. This use case of data reduction with maximum deviation that stays within the sensor noise is extremely useful for Automobile company to process data and save costs on saving huge amounts of data.

Example Use case: Autonomous Driving

Data signals from video, lidar and radar contain way more data than telematics data. To address the challenge of processing these Terabytes, Teraki has a solution which reduces the size by up to factor 10. Data can be uploaded on the DevCenter which does the data reduction based on Teraki’s market leading algorithms.

Example Use case: Driver Behaviour

Monitoring driver behaviour is getting important to prevent or predict accidents. For this particular use case the video technology comes into effect, which can detect the driver movements I.e sleepy eyes, anxiety, distraction, stress, etc. This particular use case is getting more important to reduce accidents and Teraki’s DevCenter can play an important role in quicker training and a more efficient implementation.

What to expect on DevCenter?

Currently, DevCenter is focused on supporting telematics data. However, in the next few months it will be covering additional sensors types e.g. such as video camera and lidar/radar. In one of the next DevCenter releases — scheduled in August — we are planning to allow users to upload their own video data, select how they want them to be processed and run video reduction on them. Like the telematics data, DevCenter will show all the useful metrics to the customer, so they can decide whether it would be a fit for their target use case.

Further on, DevCenter will be connected i.e. it will allow users with a simple step by step guide to connect their devices to DevCenter and to stream sensor data and run data reduction in real time.

Sign up now to use DevCenter.

Would you like to be informed on the state-of-the-art in edge processing and automotive applications, sign up for Teraki’s fortnighly Newsletter. Sign up now.

Share this Post: