RAFT: Rule-Adaptive Feedback Trainer
Implemented a rule-based system that adjusts scenario difficulty based on student performance in maritime navigation training.
- Unity
- Rule-Based System
- Python
A fixed-rule difficulty loop
RAFT adjusts scenario difficulty in maritime navigation training using fixed thresholds on student performance, a deterministic rule engine rather than a learned policy. It is the first generation in the adaptive-training line: RAFT (rule-based difficulty control) → NAT (multi-dimensional weakness adaptation) → NAT-LLM (expert-aligned LLM feedback).
Why it came first
Fixed thresholds are easy to reason about and validate, which made RAFT a useful baseline before committing to a learned, multi-parameter policy. The evaluation methodology and the Unity-based naval simulator built for RAFT carried forward unchanged into NAT and NAT-LLM. Only the difficulty-selection logic changed between generations.