A Unified Solution of Mixed Traffic Sensing, Tracking and Acceptable Active Accident Avoidance for On-Demand Automated Shuttles

OSU SMOOTH (Smart Mobile ‎Operation: OSU Transportation Hub) is an autonomous vehicle testbed that aims at providing intelligent transportation systems research for the first- and last-mile of an individual’s commute. It augments current public transportation infrastructure by providing users access to a heterogeneous network of on-demand autonomous vehicles. We use demand and usage data to provide optimized scheduling and dispatching services, which not only provide passengers access to multiple choices of transportation through different automated shuttle sizes and speeds, but also an effective, dependable and green solution for individuals with limited mobility. 

The system has been designed to maximize the replicability and scalability for future city-wide deployments. It conforms to the National ITS Architecture to maintain the benefits of possible cooperation with other subsystems of ITS. SMOOTH is based on a multi-tier B-S web architecture and a mobile-first designed web portal for access to autonomous vehicles. Utilizing a cloud based application platform, it implements the functions of three ITS subsystems：Personal Information Access, Information Service Provider and Transmit Management. Based on the communication equipment installed on the SMOOTH vehicles, the fleet of vehicles shares requests, tasks and road information. 

To ensure that the system is economically feasible, we exploit prior knowledge of the environment and take configurable approaches according to different user scenarios. The research team adopted components designed for pedestrian detection and avoidance, intersection handling, and navigation.  The pilot deployment and maintenance issues are also discussed in this presentation.
Highly dynamical movement of pedestrians and crowd traffic on sidewalks and pedestrian crossings make it difficult to predict trajectory of pedestrians. A braking system that is activated only after a pedestrian appears in front of the vehicle is not sufficient. It is crucial to analyze the situation and find the possibilities of collisions beforehand. For this purpose, an agent based Pedestrian/Crowd Motion Modeling method is developed, which simulates the social force and intension to predict a long-term pedestrian path. Furthermore, an integrated pedestrian trajectory prediction approach is designed, which combines the kinematic and dynamic aspects of the model and the agent based interactions and intention-aware models to generate a robust pedestrian trajectory prediction.  

In the first phase of tests in 2015, OSU Transportation Hub demonstrated the concept with two different types of fully-automated vehicles covering personal use and shared-use scenarios. Routes with both simple and complex traffic/pedestrian patterns are being developed as part of the next phase, in collaboration with the City of Columbus, and partially funded by the NSF EAGER program.