Semantic Structure for Intra- and Inter-Vehicle Information Aggregation and Prediction

Serial: 
UTD-2016-10-3
PI: 
Farokh Bastani and I-Ling Yen
Abstract: 

The emerging paradigm of Smart City involves several issues, including smart buildings, smart transportation, smart health-care, etc. Smart transportation is especially important since it can significantly reduce the number of roadway accidents and save thousands of lives each year by proactively preventing the occurrence of accidents.  Some important smart transportation systems include autonomous vehicle systems (AVS) and advanced driving assistance systems (ADAS).

In this project, we plan to develop a rich semantic model to capture the relevant semantic information to facilitate AVS and ADAS. The model can help improve autonomous driving in several aspects. (1) The semantic model provides a well-defined knowledge representation which describes the environment, the objects in the environment, and the situations that require attention. It can also help capture knowledge that is beyond routine sensing and analysis, such as the contextual information of the environment like weather, time, calendar events, etc., which can be used to anticipate traffic patterns and special driving cautions. Moreover, it can guide systematic sensor data analysis and provide a nice separation of low level sensor data processing tasks and high level driving decision related computation and learning. Second, information sharing can help build an extended view of the environment and facilitate coordination. It is important for each vehicle to understand the information from other vehicles and to be able to aggregate all the relevant information. A well-defined semantic model can enhance interoperability in information exchange, has the potential of reducing communication cost, and can support systematic knowledge integration. Third, learning from existing driving experiences can help with better driving decision making. The semantic model provides the means for identifying the proper learning parameters and evaluation metrics which can facilitate effective learning.

In addition to developing the semantic model, the project Co-PIs and their collaborators also plan to leverage existing AV technologies and define a semantic based computation framework to achieve enhanced driving decision derivation and learning.

University: 
University of Texas at Dallas