Code

How neural networks help explore space

How neural networks help explore space

Free Python Course ➞ Mini-course for beginners and experienced coders. 4 cool projects in the portfolio, live communication with the speaker. Click and find out what you can learn in the course.

Learn More

Data Science tools are used at all stages of the exploration of distant worlds, from the launch of launch vehicles to the control of unmanned vehicles and the analysis of the obtained data. A striking example of this is the successful landing of the Perseverance rover in February 2021. Let's consider several impressive cases of using Data Science in space technology. These technologies not only optimize processes but also increase the accuracy of scientific research, which opens new horizons for space exploration.

Launch a Rocket

Space rockets are complex structures consisting of many parts, each of which plays a vital role in a successful launch. To ensure reliability at launch, careful preparation and multi-stage testing of all components are required. The launch vehicle assembly process on the launch pad takes 30 to 40 days and involves the work of several hundred highly skilled specialists. Their professionalism and attention to detail are critical to the success of space missions.

The irony is that modern rocket technologies, considered cutting-edge, still rely on methods developed during the early days of spaceflight. Despite designers' long-standing efforts to increase payload capacity and improve fuel efficiency, the adoption of modern computer technology has been slow. As a result, many launch vehicles continue to use time-tested technical solutions developed back in the 1960s. This highlights the paradox of rocketry development, where innovation often gives way to the reliability of old methods.

Japanese designers have begun to actively change the landscape of rocketry. They were the first to widely implement artificial intelligence technologies in the development of the Epsilon launch vehicle. The use of intelligent systems allows for the automation of the launch readiness check process, significantly reducing the need for human intervention. This innovative solution helps improve the safety and efficiency of rocket launches, opening new horizons for space exploration.

Launch of Epsilon Photo: JAXA

The Japan Aerospace Exploration Agency (JAXA) has developed a robotic rocket Called "Epsilon," it can be launched with just eight people, compared to 150 highly skilled specialists required for previous models. Rocket assembly time on the launch pad has been significantly reduced to just one week. This process optimization underscores JAXA's progressive approach to aerospace technology and improves launch efficiency. Previous-generation rockets resemble outdated cars. When a malfunction occurred, only an indicator light would illuminate on the car's dashboard, leaving technicians with many questions. Finding the cause of the malfunction required significant time and effort. This complicated the work and increased the time it took to fix the problem. Modern rocket technology significantly simplifies diagnostics and improves maintenance efficiency, making the new rockets more reliable and user-friendly. "Epsilon" resembles a high-tech Japanese car, equipped with a modern onboard computer and an intelligent self-diagnostic system. This system not only alerts you to malfunctions but also provides a detailed description of the problem. Moreover, it is capable of independently troubleshooting detected problems, if this can be done using software without human intervention.

Epsilon is equipped with a high-speed data bus that connects all key components of the rocket into a single computer network. This creates the equivalent of an on-board internet. Through this network, intelligent algorithms monitor the rocket's condition. It allows for the instant analysis of large volumes of data from all components, diagnostics, and the prompt remote troubleshooting of any detected faults.

Professor Yasuhiro Morita, one of the rocket's developers, discussed the engine nozzle control system in an interview. The effectiveness of this system depends on electrical signals from numerous sensors. Nozzle deflection can cause the rocket to deviate from its intended course, jeopardizing the success of the mission. The Epsilon artificial intelligence plays a key role in monitoring sensor readings, ensuring the control system's operability. It also calibrates the nozzle's operation based on actual instrument readings before launch, significantly improving flight accuracy and operational safety.

The Epsilon rocket belongs to the light class of rockets and has a payload capacity that is approximately five times smaller than that of the Soyuz rocket. The Japan Aerospace Exploration Agency uses Epsilon to launch small satellites. To date, four successful launches have been conducted in 2013, 2016, 2018, and 2019. In the future, it is planned to modify versions of this type of rocket to send research vehicles to the Moon and Mars, which will significantly expand the capabilities of Japanese spaceflight and explore other planets in our solar system.

The Epsilon model demonstrates the effective application of Data Science in rocket engineering, despite its compact size. Similar solutions used in the Japanese development can be applied to the creation of new space launch vehicles with a high payload capacity. For example, such technologies can improve the design of SpaceX's Falcon rockets and our Angara rocket. Currently, these models lag behind the Epsilon in process automation and still rely on manual labor by maintenance personnel. The introduction of Data Science in rocket development will increase efficiency and reduce costs, making the Epsilon a significant step forward in this area.

Landing a rover

The automatic landing of a research probe on Mars is an extremely complex task. Statistics confirm this difficulty: approximately 60% of missions sent to the Red Planet have failed. The main problem is the significant distance between Earth and Mars, which causes radio signals to reach our planet with a delay of three to twenty-four minutes. This makes direct control of a spacecraft's descent from Earth impossible. Therefore, for a successful landing, probes must have a high degree of autonomy and pre-programmed algorithms.

The journey from Earth to Mars takes approximately seven months, while landing on the surface of the Red Planet lasts only seven minutes. It is at this critical moment that the risk of losing an expensive spacecraft peaks. Therefore, this phase of the mission is known as the "seven minutes of terror." Preparation for this important phase requires careful planning and advanced technology, as a successful landing is key to further exploration and exploration of Mars.

The creators of the American Perseverance rover were aware of the potential challenges associated with landing on Mars and developed solutions in advance to overcome them. One key element is the Terrain Relative Navigation (TRN) system, which provides automatic navigation of the lander during its descent to the planet's surface. This system allows the onboard computer to determine the position of the lander in real time and select a safe landing site, minimizing the risk of encountering hazardous terrain that could threaten the rover's integrity. With TRN, Perseverance can effectively explore the Martian surface, opening up new possibilities for scientific discovery and understanding of the Red Planet.

Perseverance selfie with Ingenuity in Jezero Crater on Mars on April 7, 2021. Photo: NASA

The Perseverance rover was the fifth vehicle sent by NASA to explore the Red Planet. The first missions used inflatable bladders, similar to car airbags. These bladders were filled with gas before landing, which helped soften the impact of the lander on the planet's surface. This method was first used by Soviet engineers in lunar stations. However, for the heavy Perseverance rover, which weighs over one ton and is equipped with sensitive equipment, this approach proved unacceptable.

On Curiosity's predecessor, NASA introduced an innovative landing mechanism known as the Sky Crane. This mechanism is a platform equipped with eight rocket engines. During landing, the platform separates from the lander and, using downward-facing engines, hovers several meters above the Martian surface. She then carefully lowers the rover using nylon cables. After successfully delivering the rover to the surface, the skycrane swings away, shuts off its thrusters, and falls, ensuring the safe landing of the research vehicle. This landing method has become a key milestone in the exploration of the Red Planet and has significantly increased the reliability of NASA missions.

Visualization of Curiosity's descent to the surface of Mars Illustration: NASA/JPL-Caltech

When using this method, the landing accuracy remains low. The deviation from the intended point can be two to three kilometers. Under Martian conditions, such a deviation poses a serious threat to the rover, as it could end up on rocky terrain or in a deep crater, making its recovery impossible. Therefore, it is important to develop more precise navigation and landing methods to ensure a successful mission.

Engineers working on the Perseverance rover have created the TRN automated navigation system, which has significantly improved landing accuracy. Now, the deviation during landing is no more than 40 meters. The TRN system is based on advanced intelligent image analysis algorithms, which allows it to effectively determine the location and adapt to changing conditions on the Martian surface. This technology provides more reliable and accurate navigation, which is critical for the successful completion of scientific tasks.

Scientists analyzed photographs received from orbiters to determine the proposed landing site for the rover. These images were carefully processed and loaded into the memory of the TRN control computer, which allowed the creation of a detailed map of the Martian surface. On this map, data analysts highlighted potential landing sites and designated hazard zones where the rover is at risk.

The Perseverance lander is equipped with high-speed, high-resolution cameras. As it approaches the planet's surface, the rover captures images of the surrounding terrain and compares them with a map of potentially hazardous zones stored in its onboard computer. This process ensures a safe landing and avoids undesirable situations, which is critical for a successful mission to Mars.

An intelligent algorithm analyzes the image in real time, determining the most suitable landing site. It controls the engines, adjusting the descent trajectory. As a result, the lander is directed to the safest accessible point. Then, a "sky crane" carefully lowers the rover onto the Martian surface and moves away. This high-tech system ensures a reliable and safe landing, minimizing risks and increasing the chances of mission success.

Perseverance successfully landed on Mars on February 18, 2021, and has already begun transmitting its first images. NASA released a breathtaking video of the landing process, which captures real footage rather than computer graphics. This is the first time humanity has witnessed such a detailed recording of an automated vehicle landing on the Red Planet. Perseverance's success was also made possible by advances in data science, which plays a key role in data analysis and the management of complex space missions.

Detect Water on Mars

Since the beginning of Mars exploration, scientists have tirelessly searched for water on this planet. The realization that the presence of water could indicate the possibility of life spurred interest in the Red Planet. Astronomers have long observed Martian channels through telescopes, hoping to detect signs of moisture that could confirm theories about life on Mars. Research shows that water is key to the existence of living organisms, and each new step in the study of Mars brings us closer to unraveling the mysteries of this enigmatic planet.

As spacecraft explored Mars and transmitted detailed photographs, it became clear that there is no liquid water on the planet. Previous ideas about channels turned out to be an optical illusion. These discoveries changed our understanding of the Martian surface and challenged existing theories about the possibility of life on the Red Planet. Research continues, and scientists hope to uncover new data on geological activity and possible reserves of water in other states, such as ice or water vapor.

At the beginning of the 21st century, scientists, analyzing data from automatic stations, confirmed the existence of water on Mars. However, this water is located beneath the planet's surface in the form of ice. In some regions, the permafrost layer reaches tens and even hundreds of meters in thickness. This discovery opens new prospects for studying Martian geology and searching for signs of life on this mysterious celestial body. Numerous artificial satellites are currently orbiting Mars, transmitting a vast array of images of the planet's surface to Earth. However, detecting ice accumulations in these images presents a significant challenge. Only the polar ice caps are clearly visible in the images, while the main reserves of frozen water remain hidden beneath a layer of Martian soil. Exploring these hidden resources is an important step for future missions to Mars, as the presence of water could significantly impact the possibility of colonizing and sustaining life on the Red Planet.

Data science is once again becoming an important tool for scientists. Using the Mars Odyssey spacecraft, NASA researchers were able to obtain more than 20,000 photographs of the planet's surface and create a detailed map. The Thermal Emission Imaging System (THEMIS) instrument on board Mars Odyssey measures the surface temperature of Mars by analyzing the infrared radiation emitted by the planet. This information helps scientists better understand the climate and geological processes on Mars and facilitates preparation for future missions to the Red Planet.

The temperature of individual areas on Mars can be estimated based on the amount of solar energy they receive. However, the presence of underground ice reduces this temperature. Data analysts compared the actual measured surface temperature with calculated values. If the temperature in a certain area is lower than expected, this indicates the presence of water ice beneath the surface. The difference between measured and calculated temperatures can be used to determine the depth of permafrost and the thickness of the ice layer. This research is important for understanding the climate conditions on Mars and the potential presence of water, which has implications for future missions and exploration on the planet.

NASA scientists have developed a map showing the location of subsurface ice on Mars and presented their findings in a scientific paper. This map is important for understanding water resources on the Red Planet and could play a key role in future missions to explore Mars.

White indicates where ice is located beneath the planet's surface. Image: NASA/JPL-Caltech

Search Determining the presence of frozen water on Mars based on temperature differences is a complex task. The necessary measurements and calculations are often unavailable. The non-profit startup MaritimeAI is developing a method for detecting ice beneath the planet's surface using orbital imagery. The team uses artificial neural network technologies to analyze these images, significantly improving the accuracy and efficiency of searching for water ice on Mars.

The MaritimeAI team has successfully implemented technologies for detecting shipwrecks and ice formations in satellite images of the sea surface. They are now seeking to adapt proven neural network methods to search for frozen water on Mars. These innovations have the potential to significantly improve our ability to explore the planet and open new horizons for future space missions.

To effectively analyze the images, the neural network must be pre-trained. This involves feeding it hundreds of images that indicate the location of ice beneath the surface. While photographs can be labeled manually, this process is time-consuming and tedious. Using automated labeling methods can significantly speed up data preparation and improve the accuracy of neural network training. Automating the labeling allows us to focus on more important tasks, such as optimizing models and improving their performance.

An effective method involves combining Mars images with a map of ice formations created by NASA scientists while studying temperature differences. This data makes it relatively easy to train a neural network. Once trained, the network can be fed new images of Martian soil that are not identified as containing ice. The neural network will provide highly accurate results, significantly improving our ability to study the Martian surface.

MaritimeAI published an image of the surface of Mars in which the company's neural network detected ice formations and outlined their contours as "polygons." This discovery highlights the potential of using artificial intelligence to study planetary surfaces and may facilitate further exploration of Mars.

This method could make it easier to search for frozen water reserves on Mars using orbital images in the future. This will significantly facilitate the planning of future Mars expeditions and the selection of optimal landing sites for manned stations. Colonists will not need to transport water supplies from Earth, as it can be extracted on the planet itself. The presence of water is a key factor for the existence of life.