Boosts engagement
Personalized pacing keeps advanced learners challenged and beginners supported, instead of losing either group to a fixed curve.
NSF REU Research · Summer 2025
Cybersecurity training that adapts as fast as the student learns.
ACE is a reinforcement-learning-driven VR system that reads how a learner is performing in real time — mistakes, hesitation, hint use — and reshapes the challenge around them with PPO and AIRL, instead of running everyone through a fixed script.
The Problem
One difficulty curve doesn't fit every learner.
Most VR cybersecurity labs run every student through the same fixed sequence of challenges. Advanced learners get bored and disengage; beginners get stuck without the scaffolding they need — and neither group gets training that's actually calibrated to them.
The Approach
Reading the player, adjusting the lab in real time
VR Front-End (Unity)
The VR escape room emits state, actions, and events as telemetry.
Telemetry / Features
Session features: time, mistakes, unnecessary actions, hints, completion.
AIRL
Infers a reward signal from behavior and state-action distributions.
PPO
Selects the next difficulty level and hint strategy from that reward signal.
Adaptive Hints & Difficulty
The resulting difficulty and hints feed back into the live VR session.
Why It Matters
Built for engagement, not just difficulty scaling
No manual scripting
The lab reshapes itself from live gameplay telemetry — no instructor has to hand-author difficulty tiers per learner.
Explainable by design
Every difficulty and hint decision is logged, so training stays safe, auditable, and easy to reason about after the fact.