Evidence before adjectives
“Intelligent,” “autonomous” and “human-like” are not substitutes for a test specification. Our approach starts with an environment, a movement profile, a route contract and observable outcomes. We distinguish an agent’s behavior from the quality of the world it is exploring.
This is an independent informational website, not a hosted testing service or a product benchmark. The route diagrams and trace notes are conceptual teaching examples, not recorded experimental results. Photography provides editorial context and is not evidence of a particular test execution.
Four connected investigations
The atlas separates route diagnosis, agent design, coverage and reporting. They connect in practice, but each answers a different question. A sophisticated agent cannot compensate for an undefined success condition, and a high pass rate cannot repair a missing route inventory.
Technical references and independence
We name established tools and concepts to make the guidance concrete. References do not imply affiliation, endorsement, certification or guaranteed compatibility. Useful starting points include the official documentation for Unity navigation, Unreal Engine navigation, Recast and Detour, Gymnasium, OpenTelemetry and NIST combinatorial testing.
Other references discussed across the atlas include A*, behavior trees, state machines, Unity ML-Agents, Google Research Football, Richard Sutton, Andrew Barto, Perfetto and ISO/IEC/IEEE 29119. Consult primary documentation for version-specific behavior.
Corrections and questions
If an explanation is unclear or a technical claim needs correction, identify the page and the specific statement. Include the relevant engine version or documentation when available. You can prepare a concise question with our local route-brief tool.