ALLURE: A Multi-Modal Guided Environment for Helping Children Learn to Solve a Rubik’s Cube with Automatic Solving and Interactive Explanations

Why it matters

• AI systems like DeepCubeA can solve the Rubik's Cube but do not explain their strategies in human-understandable terms.
• Educational settings need step-by-step reasoning, not only final solved states.
• Children benefit from interactive, visual, and guided learning interfaces for complex problem-solving tasks.
• There is a gap between automated planning/search systems and explainable, collaborative AI built for learning.

The system first focuses on forming the white cross (Figure 2), a standard early subgoal.

What we did

• Introduced ALLURE, a multi-modal collaborative AI system for learning to solve the Rubik's Cube.
• Extended DeepCubeA with hindsight experience replay so it can achieve arbitrary subgoals (for example, forming the white cross).
• Used inductive logic programming (Popper ILP) to extract human-understandable algorithms from solution traces.
• Translated logic-based rules into natural-language instructional explanations.
• Built a 3D interactive cube visualization and a Rasa-based chatbot for guided, explainable learning.
• Enabled collaborative use where learners can propose their own subgoals or example algorithms for refinement.

How it works

• Goal-conditioned learning: a DeepCubeA variant trained with hindsight experience replay achieves arbitrary cube subgoals.
• Example extraction: solution traces between subgoals are collected as state-action examples.
• Inductive logic programming: Popper ILP induces concise first-order logic programs that describe the move algorithms.
• Natural-language translation: induced logic rules are rendered as explanation templates for learners.
• Multi-modal interface: 3D cube visualization plus chatbot interaction (text, animation, and voice) supports guided learning.

Figure 1 shows how the system explains each move while progressing toward the white cross.

Key contributions

• Introduces a system that both learns to solve the Rubik's Cube and explains its strategies.
• Integrates deep reinforcement learning with ILP to extract human-readable algorithms.
• Enables collaborative interaction where users can propose subgoals or example algorithms.
• Provides a multi-modal explainable AI interface combining 3D visualization and NLP-driven dialogue.