AI Grid World Safe Benchmark

This repository hosts a benchmark environment designed to evaluate artificial intelligence systems under safety-constrained grid world conditions.

Agents must pursue objectives while avoiding unsafe states, hazardous paths, prohibited actions, and reward traps that can expose weaknesses in decision logic.

Benchmark Focus

• Goal achievement under constraints
• Safety vs reward tradeoffs
• Hazard avoidance
• Path planning under uncertainty
• Robustness of competing AI systems

Repository Contents

• Environment code
• Scenario maps
• Benchmark runs
• Comparative test results
• Future expansions