SocialNav-SUB
SocialNav-SUB benchmark for social scene understanding
A benchmark evaluating models on perception prediction and pedestrian action classification using VR-based navigation data from the SEAN Together dataset.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info socialnav-sub
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
3 sourced entriesEvery number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
Content & Construction: The SocialNav-SUB benchmark is designed to assess representations learned by STARS (SpatioTemporal Autoencoded Representations for Social-interaction) in Human-Robot Interaction contexts. It utilizes interaction scenarios encoded as spatiotemporal graphs where agents are nodes and interactions are edges, derived from VR-based navigation data annotated with subjective human perceptions.
Scoring & Key Findings: Performance is evaluated using linear probes on perception prediction and pedestrian action classification tasks. The benchmark demonstrates that STARS representations achieve high data efficiency, outperforming unstructured baselines in low-data regimes while capturing high-level social semantics like distinct human action intents.