Michael Saint-Guillain, a computer engineering research assistant at UCLouvain’s ICTEAM, didn’t plan on working in the fascinating realm of outer space. And yet, in 2018, while a PhD student, he participated in ‘UCL to Mars’, a unique Mars scientific expedition simulation. It convinced him to study artificial intelligence in the field of outer space.
The red planet has always fascinated us. Greatly resembling Earth, Mars is a leading candidate for comparative studies, the search for life, and possible colonisation. SpaceX CEO Elon Musk plans to send humans to Mars by 2024. Countless scientists are working to make this science fiction a reality. Among them is Dr Saint-Guillain. Last September, he defended his PhD thesis on managing vehicle routing in conditions of uncertainty. How can police patrols be optimised, for example? This has nothing to do with space—at first glance.
UCL to Mars
In May 2017 his research took an unexpected turn. ‘I applied to be in the 2018 UCL to Mars crew'. Since 2008, teams of UCLouvain PhD and other students have participated in the Mars simulation programme, based at the Mars Desert Research Station, in Utah. They carry out scientific experiments and geological work in order to optimise a future mission on Mars. The station, as well as the Mars Society, provide a unique scientific infrastructure with constraints similar to those of Mars: confined spaces, expeditions in space suits, freeze-dried food, etc.
2018 Martian expedition
Dr Saint-Guillain was part of the 2018 expedition with seven other UCLouvain students and researchers. ‘Everyone proposed an individual research project to conduct on site during the two weeks of simulation’, he says. Soil chemical analysis, drone site mapping, particle physics, etc. Dr Saint-Guillain’s project was special because it was transdisciplinary. ‘My job was to model the projects as an exercise in task scheduling in conditions of uncertainty, in order to maximise the probability that they were performed on time’. This involved operational mathematical analysis of all the experiments, with one objective: maximising the chances of mission success while taking into account available resources and the constraints and uncertainties encountered by crew members.
Calculating the most reliable schedule
In March 2018, for two weeks, the eight UCLouvain students and researchers travelled to the Utah desert to conduct their experiments. In total, 230 tasks were planned. ‘I had to find the best planning possible to combine all these tasks, taking into account all the constraints. For example, taking a soil sample requires three to five people to leave the base. This must coincide with the free moments of other researchers. The same goes for materials that shouldn’t be used at the same time by several researchers.’ To stick to the reality of space, Dr Saint-Guillain accounted for ‘uncertainty’: what’s the most reliable sequence for maximising the probability that everything goes well despite the uncertainty? On Mars, for example, uncertainty is illustrated by the window of time to communicate between the probe and Earth. If the information is not communicated at this time, the crew can only wait until the next day. A day on Mars is extremely expensive. What tasks could a crew perform while waiting for the next communication window?
Deterministic models vs. probabilistic models
‘Thanks to UCL to Mars, I was able to apply the theoretical results of my thesis to the context of a space expedition’, Dr Saint-Guillain says. After harvesting field data and continuing his experiments, he recently published his results. In his research article, he compares deterministic and probabilistic models for task planning. The context of a space mission is unique because the operations must be planned several days in advance, and complex decision chains and communication delays make last minute rescheduling impossible. According to his article, even when the probability distributions are of very poor quality, the solutions obtained by the probabilistic model largely surpass those obtained by a deterministic model. In other words, even in a context where all task durations are overestimated, the reliability of the solutions can be multiplied on average by three with the probabilistic model. The sequence of tasks succeeds in 95% of cases with the probabilistic model, against a success rate of 30% with deterministic models. ‘It’s therefore worth considering uncertainty in the context of a space mission’, he concludes.
The value of artificial intelligence
More generally, this first article allowed him to prove the value of artificial intelligence in space. The Voyager 2 probe, which explored the solar system in the late 1980s, is the best example. ‘A total of 175 experiments were planned’, Dr Saint-Guillain explains. ‘For each experiment, the probe had to be in a specific position. For each measurement, there were energy constraints. The scheduling of all these tasks occupied 30 full-time professional engineers for six months.’ The IT tool he modelled requires significantly fewer resources.
‘At the moment, astronauts don’t decide anything when they go on a mission’, he continues. ‘It would be too dangerous and cost too much. They perform what hundreds of people on Earth take great care to decide for them. For missions on the Moon or in the International Space Station, communication doesn’t take more than one second, so this kind of organisation is possible. But on Mars, there are currently only two communication windows per day of about 10 minutes. And the message takes 15 minutes to arrive on Earth and 15 minutes to return to Mars. In case of unforeseen events, astronauts travelling to Mars will have to make decisions themselves. Artificial intelligence will then support them. Thanks to a huge database, coupled with powerful problem-solving algorithms, it will suggest alternatives or decisions to astronauts.’
A very rich experience
These prospects for a space exploration revolution are exciting, as was the change of subject for Dr Saint-Guillain. ‘Going to the Utah desert with UCL to Mars did a lot for me. This student initiative completely influenced and transformed my career. For me, the thesis is not an end in itself. It's training, a search for a path. Thanks to UCL to Mars, I was able to go into a field that fascinates me, while enhancing my research.’ The 2020 crew was trained in June 2019. The next call for applications to be crew members will be in May-June 2020. Prepare for liftoff!
A glance at Michael Saint-Guilain's bio
Michael Saint-Guillain obtained his PhD in computer science in 2019 at UCLouvain (Belgium), in international co-supervision with INSA-Lyon (France). He first studied dynamic and stochastic problems of vehicle routing, then his interest quickly spread to the management of operations and decision-making in conditions of uncertainty in general. He is currently working on the application of probabilistic models for robust operations planning in the biotechnology production industry as well as for space exploration.