A future lunar outpost could have strong machines, steady power and full supply tanks, yet still face danger when tired crew members struggle with stress or teamwork. Such human problems are difficult to test before astronauts begin living far from Earth, where help may take days to arrive and a small failure can disrupt an entire mission.
Researchers have now built a virtual lunar settlement where computer-generated astronauts work, rest, consume supplies and respond to emergencies. The PLOS One study, published on May 27, 2026, presents a lunar base agent-based model designed as a benchmark for future simulations of crewed Moon missions.
Raymond Vera, Anamaria Berea and William G Kennedy of George Mason University created the model using evidence from space missions and extreme environments on Earth. Its digital crew members have different professional abilities, personality styles and emotional states. Their performance changes as they learn from repeated work, encounter lunar hazards and spend months living with the same small group.
What the lunar base model simulates
The simulation represents a theoretical Moon base on the lunar surface and the Gateway station in orbit. Astronauts follow daily schedules that include exploration, science work, repairs, exercise and sleep. They also produce or consume resources, including air, water and food, while waste collects inside the habitats.
One step in the model equals one Earth hour. A standard run covers three months, or 2,016 hours, giving the researchers enough simulated time to observe changes in crew relationships and task performance. The lunar surface appears as a grid, with each square representing about one square kilometer.
The model follows a planned settlement rather than a single early landing. Power systems, communications and life support are assumed to be operating from the start. NASA’s Artemis planning describes a long-term path toward sustained operations near the lunar south pole, where future crews could use surface habitats and mobile exploration systems.
Astronauts become digital agents
Agent-based modeling begins with individual actors that follow programmed rules. Each actor makes local choices and responds to nearby people or changing conditions. When many actors interact over time, the simulation can reveal larger patterns that are difficult to calculate with a single equation.
In the lunar model, the main actor is called an Agent_Astronaut. Each digital astronaut receives physical health values, professional skills and a personality type. Emotional conditions are represented through coping ability and tension, while technical skills cover areas such as engineering, science, spacewalking and flight operations.
The software also includes rovers, tasks and transport vehicles. Habitats store supplies and waste, while landers can move astronauts between the surface and Gateway. The researchers built the simulation in Python with Mesa, an open-source framework created for agent-based models.
How skills and stress affect each task
Every assigned activity has a difficulty threshold and requires a certain type of expertise. Exploration may depend on spacewalking skill, while maintenance calls for engineering ability. A crew member can work alone when the required skill is high enough or seek help from another astronaut.
Task success depends partly on the astronaut’s professional rating. The model then adjusts that ability according to psychological health. High tension and reduced coping capacity create an emotional penalty, which lowers the likelihood of successful work. Compatible working relationships can improve emotional conditions, while friction between personality types can raise tension.
Experience adds another influence. Repeated assembly and maintenance work follows a learning curve, allowing astronauts to become more effective over time. Skill, emotional condition and experience are multiplied together, so a change in one factor can alter the final chance of completing a task.
Personality pairings are based on the DISC system, which groups behavior into dominant, influential, steady and conscientious styles. The model assigns these types at random because detailed historical personality data for astronauts were unavailable. Certain pairings receive small teamwork benefits, while others receive penalties based on earlier research.
Inside the virtual Moon Base and Gateway
The initial case places two astronauts at the surface habitat and two aboard Gateway, supported by two rovers. The surface grid covers about 50 square kilometers around a theoretical south polar operating area. Astronauts can travel roughly one kilometer per hour during surface activities, while rovers move as far as 10 kilometers per hour.
Life inside the Moon Base and Gateway follows a daily rhythm modeled on earlier human space missions. Crew members perform scheduled jobs, exercise to protect their health and receive personal time before sleeping. Rovers explore and help build infrastructure during the day, then return for charging and routine maintenance.
Supplies place a hard limit on every mission. Astronauts consume packaged food, oxygen and water each day. Shuttle flights deliver new resources at intervals chosen by the user, with the starting model using a resupply period of about two months. The simulation also tracks local resource production and waste storage.
Transport systems carry risks as well as supplies. The shuttle can bring replacement crews from Earth, while a lunar lander can transfer people between Gateway and the surface. A user can change the timing of these movements to test whether fresh crews or rotating team combinations improve performance.
Measuring crew workload with NASA TLX
Mission success involves more than counting completed jobs. A crew could finish many tasks while suffering growing stress, which may increase the chance of later mistakes. The researchers therefore created a synthetic workload score based on the NASA Task Load Index, commonly known as NASA TLX.
NASA TLX usually asks people to rate the mental and physical demands of a task, along with time pressure, effort, frustration and their view of their own performance. The lunar simulation has no human participant answering questions, so it calculates an estimated score from the average tension level, average coping ability and total number of completed tasks.
Historical workload studies can help calibrate the synthetic score. The paper points to data from NASA spaceflight simulations, the Human Exploration Research Analog project and other isolated work environments. Psychological observations from Antarctic expeditions may also help researchers compare simulated tension with behavior recorded in a harsh and remote setting.
Testing emergencies and crew changes
Lunar crews will face events outside the daily schedule. The model gives users control over the probability of radiation episodes, moonquakes and small impact events. Lunar dust can damage equipment, while tools and life-support parts may fail during use. An accident during a resupply flight can remove both astronauts and cargo.
When a dangerous event occurs, the digital crew interrupts its normal work. Astronauts may return to the habitat, begin troubleshooting or pair workers according to engineering skill. Unexpected repairs consume time that would otherwise support exploration or science.
The study tested nine scenarios beyond the starting case. Some changed crew size or allowed astronauts to move between Gateway and the surface every two weeks. Other cases increased the frequency of surprise tasks, altered the rate of learning or extended the mission from three months to six.
One severe scenario introduced the death of an astronaut during the final third of the mission. Another stopped all crew replacements. Such tests allow planners to examine how a team responds when fewer people must carry the remaining workload while emotional pressure rises.
Results from thousands of mission runs
A Monte Carlo simulation repeats a model many times while changing selected starting values. In this case, the software varied personality types, skill levels, task requirements and environmental probabilities. Repeated runs produced ranges of possible results for workload, tension and successful task completion.
The paper’s abstract states, “Monte Carlo simulations consisting of tens of thousands of iterations show trade-offs in productivity and psychological well-being.” A scenario that produces more finished work can still place a heavier emotional burden on the crew, while a calmer team may complete fewer activities under the same mission schedule.
Extending the mission from three months to six produced the largest increase in total completed tasks, about 100 percent on average, because the astronauts had twice as much time to work. Among cases that kept the same mission duration, the favorable sensitivity case raised task completion by 27 percent. Increasing the astronaut population produced a 16 percent gain.
The scenarios also showed why raw productivity cannot serve as the only measure of mission resilience. More people can provide additional skills and labor, although each crew member also consumes supplies and adds new social relationships. Faster learning can improve repeated work, while difficult conditions can raise the synthetic workload score.
Where the first benchmark falls short
The researchers describe the model as a starting framework that can be refined when better evidence becomes available. Many inputs currently come from uniform or triangular probability ranges because direct data from permanent lunar settlements do not exist. Humanity has yet to operate a crewed base on the Moon.
Personality assignment is one major uncertainty. The model gives each astronaut an equal chance of receiving one of four DISC types, even though real astronaut groups may contain a different mix. The psychological costs and benefits assigned to certain personality pairings also rely on simplified rules.
Environmental probabilities present another challenge. Apollo records, spacecraft maintenance reports and observations of the Moon provide useful starting estimates, yet future habitats will use different equipment in new locations. Dust exposure, radiation protection and repair demands will depend on the final designs and operating procedures.
The synthetic TLX value also requires testing against human data. It combines simulated tension, coping and task output, while the traditional NASA TLX comes from people rating their own workload. Future analog missions could help researchers adjust the model until its patterns more closely match observed crew behavior.
How mission planners could use the model
Planning teams can use the framework as a virtual laboratory for situations that would be dangerous or expensive to reproduce. They could compare four-person and six-person crews, test different resupply schedules or examine how often astronauts should rotate between an orbital station and the lunar surface.
Engineers could also change the expected rate of equipment failure and see how extra maintenance affects science work. Medical and behavioral specialists could explore whether particular team structures keep tension lower during long periods of isolation. Each result would provide a range of outcomes rather than a guaranteed prediction.
As NASA and its partners prepare later Artemis missions, models of machines and supplies will need to account for the people operating them. Technical skill can improve with practice, while sleep loss, conflict and repeated emergencies can weaken performance. Agent-based simulations offer a way to place these connected pressures inside the same digital mission.
Better mission records and future lunar analog experiments could gradually strengthen the model’s human factors. With careful calibration, the framework may help planners identify crew arrangements that protect psychological health while maintaining the work needed to sustain a remote base.
