The Atlantis platform: escape that rises on demand

Opening

Mid-trial, the pool looks empty. The rat is swimming hard toward a patch of milky water that still has nothing solid under it. Then, quietly, a platform starts to rise. First a shadow, then a rim, then a place to stand. The animal climbs up, shakes, and looks round the room as if the floor of the ocean had just floated back into place.

That rising floor is the whole point. Until the animal stayed in the right area for long enough, there was no escape at all.

What the Atlantis platform is

This is the Atlantis platform, an on-demand escape in the Morris water maze. Instead of sitting fixed just under the surface from the start, it waits at the bottom. It only comes up when the animal has searched in the right place for a set time. Why invent a platform that makes escape harder? And what does that rising floor show that a fixed platform can hide?

The parts of an on-demand escape

Start with the parts. In a standard water maze, a small platform sits a centimetre or so under opaque water for the whole trial. The animal can bump into it by chance and climb up. That can look like place learning when it was only luck. It can also look like learning when the animal has merely picked up a habit, such as circling at the right distance from the wall.

The Atlantis platform flips that. At the start of a trial, the platform sits deep, out of reach. Software watches the animal's position through a camera above the pool. The animal must enter a defined target zone and stay there for a chosen dwell time. Only then is the platform released. It rises toward the surface and stops just under the waterline, still hidden, but now solid enough to stand on. Miss the zone, or leave too soon, and nothing rises. Escape has to be earned by searching in the right place.

A single trial then looks different from the classic version. The animal is released facing the wall. It swims. If it knows the place, it heads there and holds position, treading water around the spot. After the dwell time, the floor rises. If it doesn't know, it may wander, hug the wall, or cut across the pool without lingering. The trial can end without a reward.

Dwell time is a dial the experimenter can turn. A short dwell is forgiving. A longer one demands a tighter, more patient search. The zone size is another dial. A broad circle around the goal is easier than a narrow one. Together, those settings decide how precise the animal's memory has to be before the water gives it somewhere to stand.

Where Atlantis came from

The idea didn't appear from nowhere. In the mid nineteen eighties, Buresova and colleagues described an ingenious on-demand platform for the water maze. It was a collapsible, cantilevered design. It rose when a rat dwelt in a defined area for a set time. Illustrative paths suggested animals would swim to the border of the target and then tread water around the presumed spot. Escape latency grew with longer dwell times, as you'd expect when the rule gets stricter.

What that early work didn't yet settle was a head-to-head question. Did training with an on-demand platform produce clearer spatial search on a later probe than training with a conventional hidden platform?

In 1994, Spooner, Thomson, Hall, Morris and Salter published a new design in Learning and Memory. They called it the Atlantis platform, after the lost city that rises from the sea. Their version used electromagnetic control, so the raise could be crisp and repeatable. They also tied the release to video-frame tracking under software control. The animal's position decided when the platform came up. No hand on a switch. No guesswork about whether the rat had lingered long enough.

They then compared transfer-test performance after Atlantis training with performance after ordinary hidden-platform training. Transfer tests, or probe trials, are the days when the platform is absent. You watch where the animal searches. The finding that mattered was focus. At short dwell times, Atlantis-trained rats searched more tightly at the correct target location. Their training escape latencies were longer. Harder training, clearer spatial search. That's the trade the rising floor buys you.

A widely used protocol from 2006 still lists Atlantis and related on-demand designs among the useful variations of the task. Moving-platform procedures sit in the same toolbox. Teaching labs that keep Atlantis platforms on the shelf are keeping that variation alive. One public facilities page at George Mason University lists four pools. It also lists platforms for Atlantis and moving-platform paradigms, tracked with an HVS system.

What the rising floor shows

What does the Atlantis platform show that a stopwatch on a fixed platform can miss? First, it punishes non-spatial shortcuts. Circling at the right radius works less well when nothing is there to bump until you've stayed put. Sweeping the pool until you hit something fails for the same reason. The animal has to mark a place and hold it.

Second, it trains a focused search habit that survives into the probe. In a classic probe, a well-trained animal may swim straight to the old spot, find nothing, and then leave. If groups differ in how quickly they give up, whole-trial quadrant time can mis-rank them. An animal that knew the place, arrived early, and then searched elsewhere can look worse than a wanderer who drifted into the right quadrant late. Atlantis training pushes animals toward a tight search at the goal during learning itself. The probe is then less of a surprise.

Third, it softens a practical problem with repeated probes. Ordinary probes are extinction trials: no escape is possible. Run too many, and animals may learn that the platform has gone. With an on-demand design, you can keep the possibility of escape alive under rules the software enforces. That is attractive in pharmacological work that needs more than one look at memory.

Picture the numbers. Suppose the target zone is a circle around the platform position, and the dwell time is set to three seconds. An animal that crosses the zone in a straight line may score almost no dwell and get no rise. An animal that arrives and paddles in place for three seconds earns the platform. Latency goes up, because the clock includes that wait. But time near the goal, and the tightness of the search path, tell a cleaner spatial story than a lucky bump at twelve seconds.

Moving-platform versions sit in the same family of ideas. If the goal can change across days or trials, a fixed habit aimed at yesterday's spot fails. Atlantis and moving platforms both ask whether the animal can update a place memory. They are not asking it merely to repeat a swim path that once worked.

Researchers also use the rising platform as a way to separate knowing where from knowing how. An animal can learn the habit of dwelling without knowing the place, or know the place without settling long enough to raise the floor. Watching both the path into the zone and the dwell once there helps pull those skills apart. That is one reason the design still earns its keep beside the classic fixed platform.

Where it can mislead

For all that, the rising floor has limits. Longer dwell times slow learning and can frustrate animals. Stress and swimming ability still matter. An exhausted mouse that can't tread water in the zone will look as if it has no map. Vision and cue stability still matter too. If the room cues are weak, no platform design can invent a spatial signal that isn't there.

Software control adds its own demands. The camera, the tracking threshold and the zone definition have to be trustworthy. A false track that thinks the animal is in the zone when it's still at the wall will raise the platform for the wrong reason. A track that loses a dark animal against a dark pool will never raise it at all. Calibration is not a detail. It's the experiment.

And Atlantis doesn't remove the need for controls. A visible-platform session still asks whether the animal can swim and climb when the goal is easy to see. Probe analysis still needs measures that go beyond a single number. Proximity, path shape and time near the goal matter more than latency alone.

There's also a temptation to treat every rising-platform study as automatically more spatial. It isn't. Poor zone settings can collapse the method. So can dwell times so short that a brief transit counts as a stay. The design is only as strict as the parameters you choose.

Back to the rising floor

Back to that milky pool, and the platform that wasn't there until the animal earned it. The beauty of the Atlantis idea is almost theatrical. Escape arrives from below, like a small stage lift. But it arrives only for a performer who found the right mark and held it. Forty years after the first water maze papers, that trick still sharpens what the pool can ask. Not just can you get out of the water. Do you know where to wait?

You'll find the papers we mentioned, and a transcript, at water-maze.com. hvsimage.com has more on the analysis measures that turn those waits and paths into numbers.