reward-penalty-engineering

What problem does it solve?

This skill provides a process-oriented methodology for exploring, testing, and archiving reward/penalty functions in VBot navigation, enabling repeatable experimentation and knowledge retention.

Core Features & Use Cases

• Diagnose and define reward gaps by observing policy behavior and identifying concrete signals to modify.
• Hypothesize and experiment with single-variable changes, guided by defined discovery strategies, and quickly validate ideas on short training runs.
• Archive and reuse results in the reward library to prevent repeated work and share insights across projects.

Quick Start

To start, study the Exploration Cycle, reference the reward library structure, and follow the six-phase workflow (Diagnose, Hypothesize, Implement, Test, Evaluate, Archive) for iterative reward development.