Behavior Graphs: Unsupervised Graph Structures for Interpreting Robot Deployment Data
Behavior Graphs: Unsupervised Graph Structures for Interpreting Robot Deployment Data
This paper proposes behavior graphs, a method to compress robot policy rollouts into interpretable directed graphs by identifying similar behavior modes, connecting them temporally, and linking them to outcomes. In user studies across three real-world tasks, participants preferred behavior graphs over rollout videos for choosing data collection strategies, reporting 40.3% lower perceived temporal demand and 37.6% lower effort. Across 43 datasets, behavior graphs improved rank correlation between predicted and true progress by 22.9% and reduced mean absolute error by 53.5% compared to the strongest baseline.
The authors introduce behavior graphs to compress high-dimensional robot deployment rollouts into interpretable directed structures, demonstrating improved data collection strategy selection in user studies and enhanced progress prediction across 43 datasets.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
datasets
43
View original evidence
louts to choose a data collection strategy, 16 of 18 preferred behavior graphs to rollout videos, reaching comparable conclusions with 40.3% lower perceived temporal demand and 37.6% lower effort. Beyond interpretability for deployment data, behavior graphs are capable progress predictors: across 43 datasets, they improve the rank correlation between predicted progress and true temporal order by 22.9% and reduce mean
Open source S4tasks
3
View original evidence
collection decisions in. To make deployment data easier to analyze and act on, we propose behavior graphs, which compress policy rollouts into a single interpretable, directed graph. We construct the graphs by learning to identify similar behavior modes across rollouts, connecting the behaviors temporally, and finally linking them to outcomes. In a user study on three real-world tasks, where participants analyzed rol
Open source S3
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
louts to choose a data collection strategy, 16 of 18 preferred behavior graphs to rollout videos, reaching comparable conclusions with 40.3% lower perceived temporal demand and 37.6% lower effort. Beyond interpretability for deployment data, behavior graphs are capable progress predictors: across 43 datasets, they improve the rank correlation between predicted progress and true temporal order by 22.9% and reduce mean
Open source S4
collection decisions in. To make deployment data easier to analyze and act on, we propose behavior graphs, which compress policy rollouts into a single interpretable, directed graph. We construct the graphs by learning to identify similar behavior modes across rollouts, connecting the behaviors temporally, and finally linking them to outcomes. In a user study on three real-world tasks, where participants analyzed rol
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org