Recent advances in multimodal foundation models and agent systems have driven GUI agents from single-platform task execution toward cross-platform interaction. However, building multi-platform GUI agents remains challenging. On one hand, high-quality and executable cross-platform interaction trajectories are still scarce, and existing data often suffer from limited platform coverage. On the other hand, different platforms exhibit distinct interaction conventions, making joint or continual training prone to behavioral pattern mixing, platform-specific capability degradation, and catastrophic forgetting. To address these challenges, we construct Uni-GUI, a high-quality cross-platform GUI interaction dataset, and propose UI-MOPD, the first method that incorporates multi-teacher on-policy distillation into continual learning for GUI agents. UI-MOPD dynamically selects a platform-specific teacher according to the current environment and transfers platform-specific behavioral priors to a shared policy through platform-conditioned distillation, enabling adaptation to new platforms while preserving capabilities on existing ones. Experiments on OSWorld and MobileWorld show that UI-MOPD achieves task success rates of 38.2% and 12.0%, respectively, demonstrating its effectiveness in balancing cross-platform capability retention and new-platform adaptation. Project page: https://elispectre.github.io/UI-MOPD/.
Alex: Welcome to another episode of ResearchPod. Today, we're looking at a study on GUI agents—autonomous systems that can navigate screens and perform tasks on your behalf. Think of them as digital assistants that can click buttons, fill in forms, and move through apps, all on their own.
Sam: So this paper is basically asking why one of these assistants might be perfectly capable of booking a flight on a laptop, but then completely fall apart when you ask it to check a boarding pass on a mobile app?
Alex: Exactly. And the authors argue it's not simply about screen size or layout. The deeper problem is something they call "behavioral convention collapse." Here's what that means: desktop computers and mobile phones have very different unwritten rules for how you interact with them. On a desktop, you might right-click a menu or use a keyboard shortcut. On a phone, you swipe, tap, or pinch. These are genuinely different languages of interaction.
Sam: And if you train an AI on a big mixed pile of both, it tries to learn both languages at once and ends up speaking neither properly.
Alex: That's precisely it. It's like trying to learn tennis and cricket simultaneously. You might end up with a swing that's wrong for both. The model averages its expertise across platforms and loses the sharp, specific knowledge it needs for either one.
Sam: So how do the researchers actually fix that?
Alex: They propose a method called UI-MOPD. The core idea is to stop mixing everything together. Instead of one giant training pool, they create what they call "platform-specific teachers"—separate expert models, one trained deeply on desktop interaction, another on mobile. When the main agent, which they call the student, is working on a desktop task, only the desktop teacher guides it. On mobile, only the mobile teacher steps in.
Sam: So the student agent always has the right coach for the right context. It never gets contradictory advice.
Alex: Precisely. They call this "platform-conditioned teacher routing." The routing is the key mechanism—it's what prevents the agent from blending incompatible habits together. Without it, the agent might try to apply a desktop-style menu click on a mobile screen where that concept simply doesn't exist.
Sam: But here's what I'm wondering—when you teach an agent something new, there's a real risk it forgets what it already knew. Does this approach actually protect against that?
Alex: That's a well-known problem in machine learning called "catastrophic forgetting." Imagine studying hard for a maths exam, then immediately cramming for a history test—and finding that all the maths has vanished from your memory. The platform-specific teachers act as anchors. Because the student is always being corrected by an expert who specialises in the current context, it doesn't drift away from what it already knows about the other platform.
Sam: How does the system actually measure whether the student is drifting away from the teacher? How does it know when to step in?
Alex: They use a technique called "on-policy distillation." Here's the practical picture: the student attempts a real task—say, navigating to a settings menu. At each step it takes, the system compares the student's choice against what the expert teacher would have done in that exact moment. If there's a meaningful gap, the student receives a correction signal. Crucially, this only happens on the steps the student is actually taking, not on every possible action in the universe.
Sam: That's a much more targeted approach than just making the student copy the teacher wholesale.
Alex: It is. And the technical tool they use to measure that gap is called a K3 estimator. Without going into the mathematics, think of it as a very precise ruler that measures how far apart two decision-making styles are. The smaller the gap, the less correction the student needs.
Sam: Does the system keep correcting the student forever, even when it's already doing well?
Alex: No, and this is where another piece of the design becomes important. They use something called "adaptive KL masking." Once the student demonstrates it has genuinely mastered a particular part of a task—it's making the right choices consistently—the system stops applying the correction signal for that part. It essentially says: you've got this, go explore. That freedom allows the agent to improve beyond simply mimicking the teacher.
Sam: So it's a staged process. Guided correction first, then independent exploration once the foundations are solid.
Alex: That's a good way to put it. And there's one more layer worth understanding. You asked earlier about the risk of the agent becoming too narrowly trained—good at a small slice of tasks but brittle outside of them.
Sam: Right—if you're only reinforcing a narrow set of examples, doesn't the agent just learn to game those specific cases?
Alex: The authors address this directly. They use a structured reward system that grades the agent on concrete, measurable goals—things like whether it clicked in precisely the right location on screen, or whether the text it produced matched what was required. Because these rewards are tied to real, verifiable outcomes rather than just "did it look like the teacher," the agent has to develop genuine capability, not just surface-level imitation.
Sam: So the combination of platform-specific coaching, targeted correction, and measurable rewards all work together to produce an agent that's actually reliable across different environments.
Alex: That's the central claim, and the evidence the authors present supports it. By keeping platform knowledge separate during training rather than blending it together, and by giving the agent room to grow once it's demonstrated competence, the system produces more stable performance across both desktop and mobile contexts. It's a considered solution to what turns out to be a surprisingly subtle problem.
Sam: It makes you realise how much invisible structure exists in the interfaces we use every day—structure that we've absorbed so naturally we don't even notice it, but that an AI has to be carefully taught to recognise.
Alex: That's exactly right. And understanding how to teach that structure—platform by platform, step by step—is what this research is working toward. Thanks for listening to ResearchPod.