The engineering problem
Reinforcement-learning experiments require consistent contracts between observations, action selection, rollout collection, policy updates, and checkpoint state. RLBOTSLOT explores these contracts in a Rocket League simulation setting.
Learning and execution architecture
An object-centric policy represents entities in the environment, while recurrent state carries information across observations. PPO training, vectorized workers, configuration, and optional telemetry provide the experiment infrastructure.
A useful test boundary
A synthetic CPU path exercises the pipeline without requiring the full simulator or a paid telemetry service. This makes infrastructure failures easier to separate from environment integration and learning behavior.
Current evidence
The synthetic smoke check passes. Component validation currently has 46 passing and five failing tests, with limitations documented in the public repository. Full training quality, competitive performance, and rank claims remain unverified.
Current scope
Research prototype with no verified competitive rank or full-training result. Current component validation is 46 passed and five failed tests; the documented CPU synthetic smoke passes. A smoke run does not establish learning quality.