The Reinforcement Ladder

From dynamic programming to autonomous driving

Project poster — “Learn the curve.” A vintage collage of gridworld diagrams, a driver steering toy cars around a tyre, and a night racetrack, over the line “a policy learns the road by remembering what changed the future.”

07
Year — 2025

Role

  • MSc coursework — University of Edinburgh
  • Full implementation, from scratch

Five exercises, each removing an assumption of the last: exact planning with a known model, tabular learning without one, deep function approximation for continuous states, actor-critic control for continuous actions — and finally, performance engineering. The ladder ends with an enhanced DDPG agent steering a multi-lane simulated racetrack, tuned across a sweep of more than sixty trained models.

Approach

The progression is the pedagogy. Value and policy iteration solve a known Markov decision process exactly. Q-Learning and Monte Carlo abandon the model on FrozenLake’s slippery grid. DQN replaces the table with a network on CartPole and MountainCar, with a systematic comparison of exploration-decay strategies. DDPG steps into continuous action space — steering and throttle on a simulated racetrack.

The final rung is engineering: prioritised experience replay with importance-sampling correction, adaptive noise with periodic exploration bursts, orthogonal initialisation, and gradient clipping — swept across architectures and target-update rates, three seeds each.

Screenshots of the training environments: FrozenLake, CartPole, MountainCar, and the racetrack
The ladder’s rungs — FrozenLake, CartPole, MountainCar, Racetrack
Each exercise removes an assumption of the previous one — until the agent is driving.

Sweep

Algorithms implemented
8
Environments mastered
5
Models in final sweep
60+
Best mean return
374.86 ± 52.78

Findings

The sweep’s best configuration paired a larger critic with a compact actor and a faster target-update rate — larger critics improve value estimation, compact actors keep training stable, and higher τ propagates learned values sooner. Along the way, the comparative studies articulated why: why Monte Carlo is more discount-sensitive than Q-Learning, why DQN’s loss is non-monotonic, why linear exploration decay generalises across environments.

In collaboration with
  • University of Edinburgh, School of Informatics

Next case study

08 / 11