SCOUT++ Release

I’m excited to publicly release SCOUT++, a derivative multimodal human-robot instruction dataset built from the public ARL SCOUT corpus.

SCOUT++ is designed to support research on how language and visual context can be used to evaluate instruction understanding in situated robotics environments, using robot-oriented reference responses derived from the ARL SCOUT corpus.

The release includes: 12,000 GPT/model input rows 12,000 full multimodal output rows 11,980 text-only GPT output rows 12,003 aligned Navigator-view images Documentation, schema, sample data, and reproducible processing artifacts

This project grew out of my research in multimodal AI, NLP, computer vision, and human-robot interaction. My goal is to make this work easier for other researchers, students, and practitioners to inspect, build on, and evaluate. SCOUT++ is an unofficial derivative release and is not an official ARL SCOUT release. If you use it, please cite both the original ARL SCOUT dataset/paper and the SCOUT++ release. GitHub repo: https://lnkd.in/eHhFAZCq

I’d love feedback from researchers and engineers working on robotics, multimodal AI, dataset design, or instruction-following systems.