# REFORM: Recognizing F-formations for Social Robots

arXiv: https://arxiv.org/abs/2008.07668 · HTML: https://mlsanity.com/paper/2008.07668/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Aug 2020.

## Built on by (most cited first, 9 of 9)

- [A Neuro-Symbolic Approach for Enhanced Human Motion Prediction](https://mlsanity.com/paper/2304.11740/) (2023)
- [Detecting socially interacting groups using f-formation: A survey of taxonomy, methods, datasets, applications, challenges, and future research directions](https://mlsanity.com/paper/2108.06181/) (2021)
- [GROWL: Group Detection With Link Prediction](https://mlsanity.com/paper/2111.04397/) (2021)
- [Qualitative Prediction of Multi-Agent Spatial Interactions](https://mlsanity.com/paper/2307.00065/) (2023)
- [Learning Social Cost Functions for Human-Aware Path Planning](https://mlsanity.com/paper/2407.10547/) (2024)
- [F2SD: A dataset for end-to-end group detection algorithms](https://mlsanity.com/paper/2211.11001/) (2022)
- [Retail store customer behavior analysis system: Design and Implementation](https://mlsanity.com/paper/2309.03232/) (2023)
- [neuROSym: Deployment and Evaluation of a ROS-based Neuro-Symbolic Model for Human Motion Prediction](https://mlsanity.com/paper/2407.01593/) (2024)
- [Where Should I Join? Robot Group Joining via Language-Guided Goal Prediction](https://mlsanity.com/paper/2609.28467/) (2026)

## Abstract (arXiv)

Recognizing and understanding conversational groups, or F-formations, is a critical task for situated agents designed to interact with humans. F-formations contain complex structures and dynamics, yet are used intuitively by people in everyday face-to-face conversations. Prior research exploring ways of identifying F-formations has largely relied on heuristic algorithms that may not capture the rich dynamic behaviors employed by humans. We introduce REFORM (REcognize F-FORmations with Machine learning), a data-driven approach for detecting F-formations given human and agent positions and orientations. REFORM decomposes the scene into all possible pairs and then reconstructs F-formations with a voting-based scheme. We evaluated our approach across three datasets: the SALSA dataset, a newly collected human-only dataset, and a new set of acted human-robot scenarios, and found that REFORM yielded improved accuracy over a state-of-the-art F-formation detection algorithm. We also introduce symmetry and tightness as quantitative measures to characterize F-formations. Supplementary video: https://youtu.be/Fp7ETdkKvdA , Dataset available at: github.com/cu-ironlab/Babble