ConductScience VR

Starmaze in virtual reality

People find a hidden goal in the Starmaze, a maze of ten alleys first built for mice, by the landmarks around the maze or by a learned route.

ConductScience Editorial · Published November 15, 2026 · 6 references

GOALSTART
Drawing · Virtual Starmaze
Task
Find a goal with no visible marker at the end of one alley, starting from the same place on every learning trial.1,2
Strategies
People reach the goal by its place among the landmarks, a map-based strategy, or by a remembered series of turns, a route-based one.3,1
Origin
Rondi-Reig and colleagues designed the Starmaze for mice in 2006.4
Brain
In the scanner, activity in the right hippocampus predicted a map-based choice and activity in the left a route-based one.1

What the virtual Starmaze tests

The Starmaze has ten alleys, five forming a pentagon in the middle and five running out from its corners.1 Rondi-Reig and colleagues designed it for mice, to tell apart memory for places from memory for a sequence of turns.4 Iglói and colleagues developed a virtual version for people in 2009, and it has since been used with brain imaging and with older adults.3,1,2

With an allocentric, or map-based, strategy, a person finds the goal from where it lies among the landmarks around the maze.3,1 With a sequential egocentric, or route-based, strategy, a person remembers the series of body turns that leads there, such as right, left, right.3,1 The route-based strategy is a sequence: it needs the turns in a set order, unlike a string of separate turns.3 Iglói and colleagues suggest that this ordering of events may make it akin to episodic memory.3

Both strategies lead to the goal during training, so the test relies on probe trials, in which the path a person takes shows which strategy they used.1

How a session runs

A session combines learning trials, in which a person finds the way to a hidden goal, with probe trials that test how they found it.1,2 In the version by Zhang and colleagues, each pentagon alley is 16.38 virtual meters long and each outer alley 37.5 virtual meters.2 The walls are 2 virtual meters high, and the view is set 1 virtual meter above the floor.2 People move at a constant 8 virtual meters per second and turn at 95 degrees per second.2 Zhang and colleagues ran it on a computer screen, which they call a non-immersive virtual environment.2

Landmarks around the maze give people their bearings.1 Iglói and colleagues placed each landmark twice, between the ends of neighboring alleys, so that finding the goal takes the layout of the landmarks and no single one can lead the way.1 Zhang and colleagues used two similar distant landmarks for the same reason.2

  1. Practice moving

    People steer with a keypad or a joystick, moving forward and back and turning left and right.1,2 Zhang and colleagues began with joystick practice that lasted until each person felt ready, an average of 5 minutes for young adults and 15 for older adults.2

  2. Learn the way to a hidden goal

    Each learning trial starts from the same place, facing the same way, and the goal is always at the end of the same alley, with nothing to mark it.1,2 In the 2010 study by Iglói and colleagues, people started in alley 1 and the goal was at the end of alley 7.1 Fireworks go off when a person reaches the goal.1,2 Zhang and colleagues gave 9 learning trials of up to 90 seconds each, down from 11 in an earlier study, to reduce dizziness.2 Their instructions avoided any words about views or routes, so as to favor neither strategy.2

  3. Test the strategy with probe trials

    In the design by Iglói and colleagues, a probe trial starts from a new alley, alley 9 or alley 5, so the two strategies lead to different places.1 From alley 9, a person using the landmarks still heads for alley 7, while one repeating the trained turns ends in alley 5.1 Probe trials were not singled out in the instructions and ended without fireworks.1 Some training trials also ended early, so an early end did not reveal a probe.1 In the design by Zhang and colleagues, one unannounced probe trial runs in the same maze with every landmark removed, which leaves only the remembered turns to go by.2 People who then go straight to the goal, entering no other alley, are classed as egocentric strategy users.2

  4. Score the path

    Zhang and colleagues logged each person’s position and heading every 20 milliseconds, and Iglói and colleagues every 200.2,1 Distance error is the distance a person travels beyond the shortest path, so following that path exactly scores zero and every detour adds to it.2 Other measures include the number of alleys visited, the total degrees turned and speed.1,2

What the Starmaze has shown in people

Both strategies are learned in parallel

Iglói and colleagues found that people can pick up the route-based strategy from the start of training, as early as the map-based one.3 People could combine the two within a single trial, and 30% switched back and forth between them across trials on their own.3 When told to use the strategy they had not been using, every participant could switch at once.3 The authors take this as a sign that people learn both kinds of spatial knowledge in parallel.3

Left and right hippocampus

Iglói and colleagues recorded brain activity with functional MRI in 17 men as they ran the maze.1 Of the 17, 15 used both strategies at some point, and the other two repeated the trained turns on every probe trial.1 Activity in the first alley of a probe trial, before the first choice, already showed which strategy a person would use.1 Right hippocampal activity predicted the map-based path, and left hippocampal activity the route-based one.1 Activity was measured against control trials, in which people walked down two alleys joined by one turn in the same maze without landmarks.1 The authors propose that the two sides do complementary work, the right holding places and the left holding sequences in time.1

Aging

Zhang and colleagues compared 31 young adults with 30 older adults.2 The probe trial identified 17 of the young adults and 7 of the older adults as using the route-based strategy, which Zhang and colleagues call egocentric.2 Young adults had a higher average speed through the maze than older adults.2 Older adults who used the route-based strategy were as accurate as young adults, while those using other strategies took more detours and were less accurate.2 Egocentric strategy users did as well in the probe trial as in the last learning trial, and the others did worse once the landmarks were gone.2 The authors read that drop as support for the probe trial as a test of strategy.2 In older adults, keeping the egocentric strategy went with better visuo-spatial ability.2

Autism

Laidi and colleagues gave the Starmaze task to adults with autism spectrum disorder and to typically developing adults, and imaged their brains with structural and resting-state MRI.5 Adults with autism explored and learned the maze as the comparison group did, and they used and preferred the two strategies in similar ways.5 The cerebellar circuits involved in navigation did not differ in structure or function between the groups.5 The authors suggest that navigation, spatio-temporal memory and the circuits behind them are preserved in people with autism.5

How it compares with the animal versions

Rondi-Reig and colleagues drew the Starmaze design from a combination of the Morris water maze and the T-maze.4 In the mouse version, the alleys are filled with water made opaque, and the mouse swims to a platform hidden below the surface.4 The whole maze fits in a circle 204 cm across.4 Distinct wall coverings, checkered, black or white, line the inner walls of the pentagonal ring in a set order.4 Zhang and colleagues describe the human virtual star maze as similar to the rodent task.2 A person steers through the alleys on a screen, where a mouse swims through real ones.2,4

The mouse version and the version by Iglói and colleagues both train from alley 1 to a goal in alley 7.4,1 The mouse probe moves the start to alley 5, and Iglói and colleagues added a second probe start in alley 9.4,1 Mice trained for 20 days, two trials a day, with a probe test every 5 days.4 In the virtual version, people reached a plateau after five training trials.1

Normal mice solved the task with both the map-based and the route-based strategy, and 60% of control mice used different strategies across the four probe tests.4 Iglói and colleagues found the same mix in people, who learned the two strategies in parallel.3 Mice lacking NMDA receptors in hippocampal area CA1, with fewer of them in the deep layers of the cortex, were impaired at learning both.4 The hippocampus is usually given a role in map-based learning only, yet the mouse and human studies both link it to route-based learning as well.4,1

The mouse task was built to pick out more strategies than these two, including guidance and a serial strategy.4 Guidance means following a sequence of the patterned walls, and the serial strategy means visiting every outer alley in turn until the goal turns up.4 Zhang and colleagues saw the serial strategy in two older adults, who kept turning left at every junction.2

A five-path star maze for monkeys

Wirth and colleagues trained two rhesus macaques to steer with a joystick through a virtual star maze with five paths, a different design from the ten-alley Starmaze.6,1 The monkeys viewed the maze in stereo on a 152 by 114 cm screen, through shutter glasses.6 They searched for a hidden reward at the end of one path, between two landmarks, and new landmarks each day kept the layout unfamiliar.6 On average, both monkeys learned to reach the rewarded path in a dozen trials.6 About half of the hippocampal cells recorded showed spatial selectivity like that seen in rats, even though a virtual setup gives none of the balance and body signals of real movement.6 Cells tuned to place also responded to head direction, point of gaze or the task, and many fired at the sight of a single landmark.6

Build a virtual Starmaze with ConductScience VR

We write the task software in house and build each virtual Starmaze to the protocol you plan to run.

  • The alley lengths, wall height and eye height, and the landmarks around the maze
  • Learning trials with a fixed start, probe trials with new starts or no landmarks, and control trials
  • The goal, the fireworks or other reward signal, time limits, and trials that end early
  • Each trial’s path, time, speed, distance error, rotation and alleys visited, saved for analysis
  • A desktop screen with a keyboard or joystick, or a headset
  • The software alone, or a package with the headset and computer

We also build teaching versions, so students can test each other and see who follows the landmarks and who follows the turns.

Related pages

Questions

What does the virtual Starmaze measure?
It shows whether a person reaches a hidden goal by the layout of the landmarks or by a remembered series of turns.3,1 Path measures such as distance error, rotation and the alleys visited show how well the person learned the way.1,2
How does a probe trial reveal the strategy?
A probe either starts from a new alley, where the landmarks and the trained turns point to different alleys, or removes the landmarks so that only the turns still work.1,2 The path the person then takes shows which strategy they had been using.1,2
Is the virtual Starmaze the same as the mouse task?
It keeps the ten alleys of the mouse task and, in the version by Iglói and colleagues, its start and goal alleys.4,1 A mouse swims to a platform hidden under opaque water, while a person steers through the alleys on a screen to a goal with no marker.4,1,2
Which brain areas does the Starmaze engage?
In people, both sides of the hippocampus take part, the right tied to finding places and the left to following sequences of turns.1 In mice lacking NMDA receptors in hippocampal area CA1 and with fewer in the deep cortex, learning by either strategy was impaired.4
Can the virtual Starmaze run in a headset?
Yes, we build it for a headset or for a desktop screen with a keyboard or joystick.

References

  1. Iglói K, Doeller CF, Berthoz A, et al. Lateralized human hippocampal activity predicts navigation based on sequence or place memory. Proceedings of the National Academy of Sciences of the United States of America. 2010. doi:10.1073/pnas.1004243107
  2. Zhang JX, Wang L, Hou HY, et al. Age-related impairment of navigation and strategy in virtual star maze. BMC Geriatrics. 2021. doi:10.1186/s12877-021-02034-y
  3. Iglói K, Zaoui M, Berthoz A, et al. Sequential egocentric strategy is acquired as early as allocentric strategy: Parallel acquisition of these two navigation strategies. Hippocampus. 2009. doi:10.1002/hipo.20595
  4. Rondi-Reig L, Petit GH, Tobin C, et al. Impaired sequential egocentric and allocentric memories in forebrain-specific-NMDA receptor knock-out mice during a new task dissociating strategies of navigation. The Journal of Neuroscience. 2006. doi:10.1523/JNEUROSCI.3408-05.2006
  5. Laidi C, Neu N, Watilliaux A, et al. Preserved navigation abilities and spatio-temporal memory in individuals with autism spectrum disorder. Autism Research. 2023. doi:10.1002/aur.2865
  6. Wirth S, Baraduc P, Planté A, et al. Gaze-informed, task-situated representation of space in primate hippocampus during virtual navigation. PLOS Biology. 2017. doi:10.1371/journal.pbio.2001045

More virtual reality tests