← Back to experiments
RL Lab · RL-001 · in-progress
Tabular Q-Learning on CartPole-v1
A discretised tabular Q-learning agent should learn a strong CartPole policy within 500 training episodes.
Experiment setup
Environment and method.
Environment: CartPole-v1
Algorithm: Tabular Q-Learning
Status: in-progress
Outcome
Recorded result.
The first 500-episode run produced a measurable but weak policy. Mean evaluation reward was 52.0 ± 36.7 with a 0.0% success rate at the project threshold.
What this means
Interpretation and boundary.
The experiment remains open. Reproducibility, discretisation, exploration, and training stability need to be investigated before treating the method as solved.
Metrics
Recorded metrics.
- Mean evaluation reward: 52.0 ± 36.7
- Best training episode: 191
- Episodes to threshold: Not reached
- Training runs: 1
- Success rate: 0.0%
Reproducibility
Recorded configuration.
- episodes: 500
- alpha: 0.10
- gamma: 0.95
- epsilon: 1.00 → 0.01
- decay: 0.995
- bins: 10 per observation dimension
- seeds: 1 recorded run
Source and artifacts