SABER is a two-tier Bayesian trust estimator for collaborative perception between connected vehicles. The first tier keeps broad trust scores for agents and objects; cumulative-sum screening then picks out agent-object pairs with persistent omissions or unsupported reports, and the second tier checks each pair against other agents’ evidence with its own Beta state. The lowest pair score caps an agent’s trust, so many correct reports cannot dilute one targeted lie; on OPV2V it improves ROC-AUC over MATE by up to 0.427 in late fusion and raises detection rates over ROBOSAC and LUCIA by up to 96.40 and 67.07 points.
Relevance
Pooled trust has the weakness of a mean: a collaborator who is right about everything else still looks trustworthy while hiding one object. SABER is a concrete per-object fix, and the open question for robust robotics is whether it carries over from cars sharing boxes to mixed robot teams that see different things. See Byzantine fault tolerance.
Links
- arXiv
- Subfield: Security, Multi-robot systems