07 · Research
The open problems that make ASTRA real
Each is a research module in the Anoneurx Lab. None are solved yet — they are active areas of honest investigation.
Continual Learning
Learning without erasing what came before — avoiding catastrophic forgetting across long experience streams.
Task Generation
Self-generated curricula. The system should be able to produce the next useful task to try.
Memory & Recall
A memory system that can store, index and recall across session boundaries without collapsing into noise.
Self-Evaluation
Knowing how well you are doing without an external scorer — the foundation of honest, self-directed progress.
Skill Compounding
Combining learned capabilities into larger routines, instead of learning each new skill from zero.
Alignment
Steering self-modification so improvements never undermine human intent, safety or consent boundaries.
Training Efficiency
Doing more with less. Useful continual learning must be cheap enough to run constantly.
Evaluation
Honest, reproducible and boring-by-design metrics. No cherry-picked demos, no inflated claims.