LunarAI

kuber vajpayee

kuber vajpayee

Bengaluru, Karnataka

The model uses instance segmentation to classify the images into 4 classes sky, ground, small rocks, and big rocks denoted by grey, black, deep grey, and white. Github repo:- (https://github.com/kuber2001/LunarModel_Oneapi) ...learn more

Project status: Concept

Artificial Intelligence, Robotics

Intel Technologies
oneAPI

Code Samples [1]Links [1]

Overview / Usage

The development of lunar rovers presents a unique challenge for space research organizations due to the rough terrain on the moon's surface. To address this challenge, a project has been developed using artificial intelligence and machine learning techniques to prevent the rover from getting stuck in large rocks.

This project involves the use of sensors and cameras mounted on the rover to detect and classify obstacles on the surface. The machine learning algorithm is trained on data collected during testing to recognize patterns in the terrain and predict potential hazards.

The algorithm can then adjust the rover's movement to avoid obstacles and prevent it from getting stuck. This project will be an important tool for space research organizations as it will improve the safety and efficiency of lunar rover missions, allowing for more exploration and discovery on the moon's surface.

Methodology / Approach

To develop this model we use the deep learning algorithms of Convolutional Neural Network that has multiple layers deployed in a U-net Architecture.

The Architecture is employed using Tensorflow framework and additionally trying to use the Functional programming methodology to create the entire architecture and pipeline.

Our project lunar api as established in the documentation uses Segmentation model that is developed using CNN in a U-net Architecture. As a result based on the huge dataset we are training the model on, it requires optimization.

The average time for training of the model was approx. 50 mins to 1 hr in our local systems. Moreover in some local system it failed to execute at all. In comes the oneapi, using the devcloud access provided to us by Shriram Vasudevan sir and with the guidance of other mentors such as Arun, Akshay and Joel, we learnt how to migrate our code to devcloud and use the Intel Analytics toolkit.

The most prominent of them all, was onednn, which was optimized in tensorflow 11.0 version to get the required training output in mere 35 mins. It boosted the speed significantly, additionally we are trying to use openvino toolkit to make it more faster and perhaps reducing the training time to 20 mins.

Technologies Used

Python, Django,OneDnn

Repository

https://github.com/kuber2001/LunarModel_Oneapi

Collaborators

3 Results

3 Results

Comments (0)