BehaviorFL-MT: Federated Multi-Task Learning under Unequal Vehicle Sensors and Partial Supervision

Connected vehicles differ not only in what they observe but also in which targets they can supervise: a phone-grade client may expose sparse GPS, whereas an OBD-equipped vehicle can support a telemetry-derived powertrain-state composite. We introduce BehaviorFL-MT, a federated multi-task formulation that predicts destination region and this composite from an early recorded-sample prefix. Two explicit masks separate missing inputs from missing targets, structured OBD dropout exposes the state head to equipped and GPS-only views, and discovery-aware participation increases rare-task exposure without prior client-label metadata. Destination prediction is anchored by a fixed prefix-geometry prior with a small bounded learned correction. On 102,843 VED, GeoLife, and T-Drive trips with client-disjoint evaluation, discovery-aware participation raises natural/GPS-only state AUPRC from $0.512/0.452$ (the best uniform-participation federated baseline) to $0.613/0.552$. An oracle with complete eligibility metadata from round 1 reaches $0.613/0.550$, indicating that online discovery recovers essentially all of the observed oracle gain despite discovering only 114-129 of 188 eligible clients by round 60. Destination HR@5 remains $0.748$ versus $0.751$, with a macro-F1 tradeoff from $0.339$ to $0.325$. The resulting study isolates how partial task supervision, unequal sensors, and cohort composition interact in federated behavioral learning.