Skip to content
Home » Panasonic Advanced Technology Co., Ltd. Started joint research with JAXA on prototype driving support AI for lunar exploration rover

Panasonic Advanced Technology Co., Ltd. Started joint research with JAXA on prototype driving support AI for lunar exploration rover

[Panasonic Advanced Technology Co., Ltd.] Started joint research with JAXA on prototype driving support AI for lunar exploration rover
*View in browser* *Panasonic Advanced Technology Co., Ltd.*
Press release: June 4, 2024
**
Started joint research with JAXA on prototype driving support AI for lunar exploration rover
*Using know-how in developing sensing technology for rough terrain environments, we have begun prototyping a driving support system for the lunar exploration rover*
* Panasonic
Advanced Technology Co., Ltd. will begin joint research with the Japan Aerospace Exploration Agency (JAXA) on “prototype production of driving support AI for lunar exploration rovers using CG images and a small amount of data.” *

This research is based on the Space Exploration Innovation Hub promoted by the Japan Aerospace Exploration Agency (JAXA).
Conducted at the “Open Innovation Hub for Expanding Humanity’s Survival Sphere and Area of ​​Activity by Developing the Solar System Frontier” (* *1)*
Based on the research results, we developed a prototype driving support system that recognizes the lunar surface environment by improving the ability of a stereo camera to detect rocks and craters as obstacles that impede the safe movement of the lunar surface exploration rover. Masu.
                              @JAXA Space Exploration Experiment Building Operation screen image (under development)
In recent years, object detection models based on deep learning have been used in a variety of situations, but their development usually requires a large amount of training data, making the development cost prohibitive. In addition, when adapting to dangerous scenes such as disaster sites or places where data cannot be easily collected such as the space environment, it is not possible to prepare a sufficient amount of training data, and the performance that AI requires is insufficient. There is also the problem that it is difficult to demonstrate.
There is a conventional deep learning method using high-quality rendered CG images, but since there is quite a gap between CG images and real data, it has been difficult to create highly accurate deep learning models for CG images. Even so, in a real production environment there will be a domain shift problem that will reduce accuracy.
In this research, we used a simulator to generate a large amount of CG images simulating the lunar surface environment and constructed them as source domains, and compared them to a small amount of actual photographed data of the lunar surface* (*2)*
By using adversarial learning and semi-supervised learning as the target domain and performing domain adaptation, we are testing a deep learning method for object detection that does not reduce accuracy even with a small amount of supervised data.
This research and development will enable the development of low-cost and highly accurate AI, and will allow AI systems to be applied to business areas that have not been widely used due to the difficulty of obtaining training data even on the ground. You can expect it to become.
In the research and development of environment recognition technology assuming the lunar surface, an important element is the construction of a virtual environment that reproduces the lunar surface. In this research and development, we will use 3D data of the lunar south pole published by NASA (*
*3)* is imported into the game development engine Unity, and from there it is used to create the expected travel task locations for the exploration rover in the Artemis project (* *4)*
We are building a 3D simulation environment that simulates sunlight shining in from the side. Please note that the published topographical data has a low resolution of 5m/pixel and lacks information on small craters and rocks less than 10m in size that would impede the movement of the lunar surface exploration rover.
*5)* Virtual craters and rocks are placed as obstacles as a reference. Our company has the know-how and track record of sensing development, model-based development, and simulation development in in-vehicle product development, and has built a common PF for autonomous driving development that can be evaluated in both a simulator environment and an actual machine/actual environment. .
system configuration diagram
We are conducting demonstration experiments using a small rover in a simulated lunar surface environment at the Space Exploration Field of the JAXA Space Exploration Experiment Module. Assuming the Antarctic region, we are evaluating robustness in a lighting environment that simulates sunlight shining in from the side, and are working to improve performance.
                            @JAXA Space Exploration Experiment Building We will continue to further improve performance, adapt to use cases, and promote research and development in order to establish
technologies that will realize a variety of autonomous mobility. * [Note] *
(*1): “Trial of deep learning method for object detection using CG composite images for small amount of data”
  https://www.ihub-tansa.jaxa.jp/topics/RFP_announcement8.html Initiative introduction video (Youtube):
  https://www.youtube.com/watch?v=25eDuX1d6zY
interview:
  https://www.ihub-tansa.jaxa.jp/topics/interview_article_panasonic.html (*2): Lunar surface exploration rover public image data (Chang’e 3, 4)   https://moon.bao.ac.cn/
(*3):South Pole Landing Site LOLA DEMs
  https://pgda.gsfc.nasa.gov/data/LOLA_5mpp/
(*4):Traverse into crater/PSR:LUNAR SURFACE DATA BOOK Artemis Campaign Development ACD-50044 @NASA
(*5):SLS-SPEC-159: Cross-Program Design Specification for Natural Environments (DSNE) Revision I

Mission logo: Panasonic Advanced technology Development challenges the moon. * [Inquiries] *
Panasonic Advanced Technology Co., Ltd.
https://adtsd.jpn.panasonic.com/contact/
contact.pad@ml.jp.panasonic.com

* About Panasonic Advanced Technology *
Panasonic Advanced Technology Co., Ltd. is an affiliated company of Panasonic Holdings Co., Ltd., and is a company whose main business is software/system development.
In 2012, ASIL D (Automotive Safety Integrity
Level: Certified with the highest automotive safety level (D). We are involved in a wide range of mobility developments, including the development of in-vehicle ECUs that support functional safety (automatic parking ECUs/transmission control ECUs, etc.), autonomous driving of construction machinery, and goods transportation robots. In addition to mobility, we are also working on technology development and business development in the housing, robotics, and security fields.
https://adtsd.jpn.panasonic.com/
*About details about this release*
https://prtimes.jp/main/html/rd/p/000000008.000109020.html

*Download press release materials*
https://prtimes.jp/im/action.php?run=html&page=releaseimage&company_id=109020&release_id=8


Unsubscribe HTML email

Copyright © PR TIMES Corporation All Rights Reserved.

%d