Quantitative Reward Monitors for Reinforcement Learning

This repository provides the reproducible code for our paper "Expressive Temporal Specifications for Reward Monitoring" for AAAI 2026.

Cite

To be added

Setup instructions

1. Clone with git git clone --recurse-submodules --shallow-submodules https://github.com/nightly/quantitative-reward-monitoring
2. Ensure GraphViz is installed on your system (used for export visualisations)
3. Use requirements.txt with pip pip install -r requirements.txt or, preferably uv sync if you have Astral's uv installed

Code structure

Credits

Original code & ideas contributed by:

• Jialue Xu
• Alexander Rader