Action Planning for the collision Avoidance System Using Neural Networks
Arain, M.A., Tribe, R., An, P.E. and Harris, C.J. (1993) Action Planning for the collision Avoidance System Using Neural Networks. Intelligent Vehicle Symposium , 119-124.
Full text not available from this repository.
An understanding of the scenario in complex traffic situations is essential in order to give an early warning, or in an autonomous system, to intervene in the urban or motorway environment. A collision avoidance system needs both to predict possible collisions or hazards and to plan a less hazardous move in a critical situation. A crucial factor in the success of the system is the use of a priori knowledge. The classical problem with a knowledge-based decision making system is the acquisition and representation of the knowledge. It is diffcult to design and develop a system for real time auto-piloting in varied traffic environments. Neural networks are ideally suited for applications where a large training set is available because they can apply human decision making criteria in different situations. The learning processes encapsulate a wide variety of drivers' reactions to various scenarios. Neural networks' abilities to generalise their training to new scenarios in the light of driving experience and to make emotion-free decisions leads to a system that is adaptive and closely which resembles human action strategy. Recognition of a scenario is achieved by acquiring data about a scene from a variety of sensors. Visual data is preprocessed and features are extracted using a real-time image processing system, while microwave radar provides obstacle information and distances. This paper described an early warning system and suggests possible responses to various traffic situations. The paper focuses on various learning algorithms for decision making which is based on the current model and immediate history only. It would help if we could always recognise the dominant threat at every instant and avoid it by either slowing down or changing direction. In our analysis of situations using neural networks, the test cases show that reasonably such behaviour can be generated. In order to validate the auto pilot it is tested in parallel with expert drivers to assess the drivers' action in a number of scenarios. The network's intervention control is verified by independent observers. The intervention strategies are based on a number of rules by which an intervention controller is trained to generate various actions. These rules are fine tuned on-line to achieve reliable and repeatable actions.
|Item Type:||Conference or Workshop Item (UNSPECIFIED)|
|Additional Information:||Address: Tokyo, Japan|
|Divisions:||Faculty of Physical Sciences and Engineering > Electronics and Computer Science > Comms, Signal Processing & Control
|Date Deposited:||04 May 1999|
|Last Modified:||27 Mar 2014 19:51|
|Further Information:||Google Scholar|
|RDF:||RDF+N-Triples, RDF+N3, RDF+XML, Browse.|
Actions (login required)