Can We Create A Self-Driving Spacecraft Using AI?

Driverless cars are the future of transport and pioneers in this technology like Tesla have spent years researching, creating and fine-tuning the model needed to make a 'perfect' AI that can drive while you sit back and relax. Along with the obvious complexity of the algorithms needed to make a driverless car lies another challenge when making a self-driving vehicle - data. To train these artificial intelligence models, lots and lots of data needs to be available to be able to train and test the car, so it will actually drive and not crash into the first lamppost the computer 'sees'. This is why companies have spent more than a decade collecting this crucial data - the fuel of any AI - from hundreds if not thousands of cars that are equipped with sensors and cameras.

So what about space travel?

There is no database that exists where companies are able to use it to train a spacecraft, but that isn't surprising. Unlike driving a car down the road, you can't just 'drive' a rocket to space. It's a process that takes months if not years of preparation, a considerable amount of money and a lot of fuel. A rocket that can take people to outer space can take 5 years to make and that's just the building process not including the time taken to do research and plan everything out. SpaceX's Saturn V rocket cost the company a nice round figure of $185 million. Also, think about the Space Shuttle that weighed 165,000 pounds when it was empty and the fact that it needs to hold around 3.3 million pounds in fuel so that it could brave through a rocket's number one enemy - gravity.

The Space Shuttle (Source: NASA)

The Space Shuttle (Source: NASA)

What do scientists do to overcome this challenge? Simulations. Scientists can simulate how a spacecraft would act and travel in the cosmos, which is a huge step forward in terms of coming closer to the answer to how we can find a way to get the necessary data, but all of these models and simulations still don't have the same impact as real-world data. So what do we do now? We wait.

When there's a demand for something new, it won't be too long until there's a supply, so when it becomes possible to collect this essential data, companies will start to collect and sell this data in order for it to be used to train these driverless spacecraft AIs. There is already a similar situation happening in the world of ground-based driverless vehicles and automated machinery.

However, data isn't the only problem when trying to create a self-driving spacecraft. When an AI comes across something new like in a case where an algorithm trained using pictures of apples sees an orange, it will freak out and not know what to do. It's fine if it happens to the AI that you made for fun, but it can be disastrous if it happens to a spacecraft as it could crash wasting millions of dollars or possibly endanger humans. In a universe that we've explored about 4% of how can we be sure that something new won't come up to 'freak out' and confuse the spacecraft's AI?

The Hera Mission (Source: ESA)

The Hera Mission (Source: ESA)

With these challenges in mind, you might think what the point is of investing so much time and money into spacecraft and space travel other than just for scientific discovery, but there is a big one that can directly impact our planet. Spacecrafts powered with AI might be able to recognise a threat, such as an incoming asteroid, and proceed to take action in order to put the Earth out of harm's way like the proposed Hera mission that aims to visit the asteroid Didymos. National and international security is a key part of the reason why self-driving and AI-powered spacecraft may be of use in the future. Also, satellites can use AI in order to learn and understand data in the outer stretches of space that we cannot see yet. Needless to say, there are many potential benefits to these type of spacecraft and will most likely be developed in the future, but the question is, when do we get there?