This is the pre-proceedings for the RLC 2026. You may expect minor changes.
Reinforcement Learning Journal, vol. 7, 2026, pp. TBD.
Will be presented at the Reinforcement Learning Conference (RLC), Montréal, Quebec, Canada, August 15–17, 2026.
Keywords: Deep Reinforcement Learning, Benchmarks, Evaluation, Procedural Generation,
We present Procedurally Generated Grid-Based Traffic Gym (PGTG), a Gymnasium-compliant, feature-rich, extensible, and customizable reinforcement learning benchmark. PGTG requires an agent to navigate procedurally generated tracks while having to manage momentum and being exposed to various stochastic obstacles and the unpredictable behavior of traffic participants. By providing fine-grained control over the map generation, traffic rules, obstacles, and the observation and reward functions, PGTG is a versatile benchmark for various reinforcement learning subfields, such as exploration, safety, and generalization. For instance, PGTG allows decoupling safety from exploration by including subgoals with intermediate rewards. In this paper, we detail the benchmark’s modular design philosophy and provide a comprehensive empirical evaluation using state-of-the-art deep reinforcement learning baselines, highlighting PGTG’s complexity and utility for future research. Our code is available at https://github.com/neuro-mechanistic-modeling/pgtg.
Joshua Meyer, Felix Maurice Kuntz, Verena Wolf, Jörg Hoffmann, and Timo P. Gros. "PGTG: Procedurally Generated Grid-Based Traffic Gym." Reinforcement Learning Journal, vol. 7, 2026, pp. TBD.
BibTeX:@article{meyer2026pgtg,
title={PGTG: Procedurally Generated Grid-Based Traffic Gym},
author={Joshua Meyer and Felix Maurice Kuntz and Verena Wolf and Jörg Hoffmann and Timo P. Gros},
journal={Reinforcement Learning Journal},
volume={7},
pages={},
year={2026}
}