reward-penalty-engineering - Explore and archive VBot navigation rewards
Provides a systematic methodology to diagnose, formulate, test, evaluate, and archive reward and penalty functions for VBot quadruped navigation.
Tags
Updated: 2026-10-15Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Diagnose reward-related behaviors
- Formulate testable hypotheses
- Implement minimal reward changes
- Run AutoML reward tests
- Evaluate experiment outcomes
- Archive reward findings
Inputs
- Observed robot behaviors
- Reward implementation and documentation
- Reward hypotheses
- Training configuration
- Reward search spaces
- Archived reward library
Outputs
- Modified reward configuration or code
- Training behavior and metrics
- AutoML comparison reports
- Archived reward findings
Requirements
- VBot navigation codebase
- Simulation and training environment
- Access to reward configuration files
- Write access to the reward library