Klepsydra AI online Demo

Overview of the online demo

This online demo shows Klepsydra AI in action and the power of our.algorithm to process datain real time. Specifically, this demo demonstrates the pose determination of an asteroid using artificial intelligence algorithms.

The use of AI is very attractive to the Space sector, but its intensive power consumption currently prevents its use in real missions. Klepsydra solves this challenge for the Space sector with the advanced algorithms and low power consumption of Klepsydra AI.

The demo presents the output of three different products: Klepsydra AI, TensorFlowLite and OpenCV. During the demo, the output of these three products shows in real time the video streaming of processed images, CPU usage and data throughput. This allows users to visually observe the different performance of all three products and more importantly, the power of Klepsydra AI.

Artificial Intelligence

Artificial intelligence (AI) is becoming instrumental in solving all kind of real life problems: computer vision, face recognition, robotics applications, health care, etc. However, there are some challenges related to computing power requirements when running AI models, especially in the case of edge computing, i.e, in small devices like car computers, IoT, etc.

One challenge is AI model performance in small computers, which is usually very poor. Current solutions for AI on the edge experonce limitations in data processing volume such as data losses, long latencies and CPU saturation.

Klepsydra has solved these problems by developing an inference engine, based on our advanced software algorithms, that executes AI models more efficiently. Consequently, Klepsydra processes 2x more data with 50% less CPU than the leading competition, as you will see in the demo.

Credits

This demo is a joint effort with several parties. The original images, out of which the synthetic images in the demo are made, come from the ESA Rosetta mission, the AI algorithm was developed in collaboration with the Universidad Carlos III, and in particular by the research group run by Manuel Sanjurjo. Without them, this demo would have not been possible.

The Klepsydra AI online demo

Our technical team has developed this demo in order to show the Klepsydra AI inference product capabilities, for Asteroid pose estimation. The demo compares the performance of three different products: Klepsydra AI and the main market leaders, OpenCV and TensorFlowLite.

During the demo, each product is running on the same, but separate, small computer; described below
The technical specifications of the system running the demo are as follows:

AI model: Multibranch DNN based on AlexNet for 224x224x3 images
Processor: Intel 3.3GHz Dual core
Data rate: 10 Frames per second (FPS)
Products: Klepsydra AI, Tensor Flow Lite, Open CV
OS: Ubuntu 20.04

How to run the online demo

Step 1

The following link has to be used to start the DEMO:

• For x86: https://space.demo.klepsydra.com/asteroid-admin/
• For ARM: https://spacearm.demo.klepsydra.com/admin/

Firstly, go to the link and wait 5 seconds for the table to show the status of the demo (active or inactive). If they are in green, the demo is running correctly. However, if they are red, the demo has stopped or finished (it lasts 20 mins). If the demo is not active (RED), then press the start button and wait 5 seconds for the demo to start:

Step 2

If all four boxes are green, including the text OK, it means the demo was started correctly. After the demo is started and processes are running, press the “go to demo” button:

This button takes you to an instance of NASA OpenMCT. OpenMCT is a real time monitoring web application that Klepsydra uses to access performance metrics. On the left, there is a menu. Select “Klepsydra demo”, which should be at the bottom:

Step 3

Once selected, a list of boxes will appear on the right. Click on the one named “asteroid demo”:

Understanding the demo

Once pressed, you will see the actual demo page. On this page you will see the pose estimation algorithm running in real time for the three products. This DEMO is in real time and it lasts 20 minutes. After this time, it will stop automatically. This page contains 3 elements:

• Data throughput chart (top left). This shows the speed in Hz at which Klepsydra (blue), TensorFlow(red) and OpenCV(yellow) can process images.
• CPU consumption chart (bottom left). Percentage of use of the computer by the three above mentioned solutions.
• Video streaming (right). There are three boxes here, one per solution (Klepsydra at the top, TensorFlow in the middle and OpenCV at the bottom). These images depict the asteroid position based on data processed by the three products.

Conclusions

The results of the demo show the power of Klepsydra AI very clearly:

• More deterministic: Klepsydra CPU and throughput remain flat throughout the duration.
• Zero data loss. TensorFlow and OpenCV cannot process the data and display the images at 10FPS. Klepsydra has ZERO DATA LOSS!
• Klepsydra CPU usage is 50% less than TensorFlow’s.
• Klepsydra Throghput is 2x more than TensorFlow’s and even more with respect to OpenCV!