Rule-VLN is a rule-compliant outdoor vision-and-language navigation benchmark built on the Touchdown / StreetLearn urban navigation environment. It studies whether navigation agents can follow language instructions while also complying with semantic traffic rules, such as regulatory signs that prohibit otherwise reachable movements.
This dataset accompanies the paper:
Rule-VLN: Bridging Perception and Compliance via Semantic Reasoning and Geometric… See the full description on the dataset page:
https://huggingface.co/datasets/jeffry77/Rule-VLN.
