Premium
Collision Avoidance Using a Cerebellar Model Arithmetic Computer Neural Network
Author(s) -
Eskandarian Azim,
Thiriez Stephane
Publication year - 1998
Publication title -
computer‐aided civil and infrastructure engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.773
H-Index - 82
eISSN - 1467-8667
pISSN - 1093-9687
DOI - 10.1111/0885-9507.00109
Subject(s) - computer science , artificial neural network , task (project management) , controller (irrigation) , state (computer science) , function (biology) , control engineering , artificial intelligence , algorithm , engineering , agronomy , systems engineering , evolutionary biology , biology
Avoiding collisions with obstacles in a clustered environment is a difficult task for autonomous vehicles. Deterministic algorithms cannot address all scenarios encountered and may fail to perform in dynamically changing environments. Neural networks, owing to their ability to map complex relationships between multiple input‐output patterns, can learn the task of maneuvering around and in‐between obstacles to reach a goal state. The Cerebellar Model Arithmetic Computer (CMAC) neural network in particular is based on a model of human sensory motor responses and can efficiently model responsive control actions. A CMAC neural network controller was developed and examined, in simulation, for its suitability to capture a driver's function of steering and braking. The performance of the controller was tested in a simulation of a moving platform (vehicle) encountering obstacles of various shapes, whereas the CMAC was trained only with limited shapes and scenarios. Preliminary simulation results have shown the CMAC's ability to successfully generalize its learned patterns to avoid obstacles after only a few training sessions. The CMAC output is generated in a computationally efficient manner with physically and economically realizable memory sizes. Therefore, real‐time hardware implementation of the controller is feasible. This research demonstrates that the method has the ability to accommodate more complex scenarios.