RL Lab / learning
Learning Progression
A structured record of learning activities connected to implementation, experiments, evaluation, and next engineering questions.
From fundamentals to experiments
Learning entries are most useful when they connect a concept to a working implementation, a measured outcome, or a technical question worth testing.
- Q-Learning Navigation in a 4×4 GridWorld: GridWorld made the tabular update concrete: Q-values encode long-term action value, obstacles create explicit negative feedback, and the greedy policy can be inspected directly. This experiment is deliberately a bridge into deeper value-function methods rather than a claim of broad Deep RL capability.
- Tabular Q-Learning on CartPole-v1: The experiment remains open. Reproducibility, discretisation, exploration, and training stability need to be investigated before treating the method as solved.
- Deep RL Lunar Lander: This establishes the first completed Deep RL environment in the public learning log. The replay demonstrates learned behavior; standardized reward benchmarks and multi-seed evaluation are intentionally not claimed until they are logged.