IRLLABIndependent Reinforcement Learning Lab

Independent RL Lab.

Experiments in sequential decision making under uncertainty. I build controlled environments, solve what I can exactly, and test where learned behavior changes.

Latest project
Interactive build 004
Status
Human and agent flight comparison

Interactive build 004

A trained policy, and the same controls in your hands

Fly a lunar lander, watch a policy-gradient agent, and compare their flights from the same starting conditions. An implementation of PPO with shared browser and training physics.

PPO · Browser-native physics · Interactive flight comparison

Take the controls
Two flight paths and a lunar lander approaching the same landing pad.

Play in your browser · Watch the agent · Replay the same landing

METHODS

Methods used here.

  1. 01
    Exact references.

    When the transition model is known, I calculate a dynamic-programming reference before training.

  2. 02
    Matched comparisons.

    Methods use the same interaction budgets, seed panels, and held-out evaluation conditions.

  3. 03
    Pilot and full-run roles.

    Pilot runs calibrate the protocol. Only the declared full runs support reported results.

  4. 04
    Exceptions and failures.

    Instability, unresolved actions, failed trials, and results that complicate the summary remain visible.

Read the methods

STUDIES

Study log.

Interactive build 004Learning to landOctober 2026PlayableClosed pilot 003A replay-fault pilot that stoppedSeptember 2026ClosedResearch note 002The timing of bad luckAugust 2026PublishedCalibration 001Shortcut or shelter?August 2026Published