This repository contains the STT-Arena benchmark, 227 interactive tasks for evaluating LLM agents under spatio‑temporal dynamics.We also release 2,212 refined SFT trajectories and verifiable RL tasks used to train our STT-Agent.
STT‑Arena tests adaptive replanning when a sudden spatio‑temporal change invalidates an ongoing plan.Tasks cover 9 conflict types (temporal, spatial, spatio‑temporal) and 4 difficulty… See the full description on the dataset page:
https://huggingface.co/datasets/Miaow-Lab/STT-Arena.
