SCERL
Repository:
Description
This repository contains the source code and data for our paper SCERL: A Text-based Safety Benchmark for Reinforcement Learning Problems. SCERL is a text-based environment for reinforcement learning agents that:
- provides a framework for genereting safety problems representing key safety challenges such as negative side effect, scalable oversight and safe exploration
- includes a pre-generated set of text-based games with safety constraints in order to spoor research in safe and text-based reinforcement learning (see dataset/safety_games).
Main Contributors
Lan Hoang, Shivam Ratnakar, Nicolas Galichet, Akifumi Wachi, Keerthiram Murugesan, Songtao Lu, Mattia Atzeni, Michael Katz, Subhajit Chaudhury