Multi-Agent Framework for Transit Passenger Load Estimation
A novel framework has been introduced for estimating passenger loads in transit systems, characterized by a closed-loop, state-centric, and multi-agent approach utilizing diverse data streams. This method tackles issues like incremental counting inaccuracies, conflicting evidence, and sensor reliability that varies with context. It ensures physical feasibility throughout the process, dynamically adjusts trust levels among various evidence sources, and incorporates physics-based violation residuals into training to enhance robustness. The architecture features a unified stop-event backbone along with a coupled Perception-Physical-Fusion loop for processing at each stop. This research can be found on arXiv with the ID 2605.19834.
Key facts
- Framework is closed-loop, state-centric, and multi-agent.
- Designed for passenger load estimation from heterogeneous data streams.
- Addresses incremental count errors, evidence conflicts, and context-dependent sensor reliability.
- Enforces physical feasibility at every step.
- Dynamically allocates trust among evidence sources.
- Feeds physics-derived violation residuals back into training.
- Architecture includes unified stop-event backbone and Perception-Physical-Fusion loop.
- Published on arXiv with ID 2605.19834.
Entities
Institutions
- arXiv