IEEE Wireless Communications • December 2014 41 INTRODUCTION Vehicular ad hoc networks (VANETs) are spe- cial types of networks in which movement of nodes/vehicles on the road is constrained by road conditions such as density and velocity of vehicles. These types of networks are primarily used to provide safety and comfort to passengers on the road. The communication among the vehicles may be peer-to-peer (P2P), which is called vehicle-to-vehicle (V2V), or with infra- structure support, called vehicle-to-infrastructure (V2I) [1]. Short-range communication among vehicles may be performed by dedicated short - range communication (DSRC) using the IEEE 802.11p protocol, while for long-range transmis- sions, WiMAX or radio communication may be used [1]. Vehicles may work together for infor- mation sharing so that adaptive decisions about the selection of particular route can be made [1]. Also, vehicles generally access various resources from the nearest access points, called roadside units (RSUs), which are deployed along the road. These RSUs may be located in such a manner that coverage and connectivity are always maintained. As vehicles move at high speed on the road, data dissemination to all the vehicles on the road is one of the major issues requiring special atten- tion. The situation becomes more complicated if, in addition to the variation in the speed of vehi- cles, the density of vehicles on the road also changes. Hence, in view of the above, a special- ized approach is required which takes these issues into consideration so that adequate steps can be taken in advance to overcome this situation [1]. There are a number of applications which require reliable transmission, minimum end-to- end delay (E2ED), and scalability in VANETs. Reliable transmission concerns the ratio of the number of packets received to the total number of packets sent from the source. E2ED concerns the occurrence of delay in transmission of infor- mation from one end to another, and scalability concerns the overhead occurrence during the transmission. However, due to the high mobility and constant topological changes, it is challeng- ing to design an effective solution within the constraints of high packet delivery ratio (PDR) and low E2ED [2–4]. To increase the reliability of message dissemi- nation of vehicles in VANETs, minimum span- ning trees need to be constructed and have been used in many engineering applications [2–4]. In this context, vehicles may be grouped together to form a cluster for information sharing and mes- sage passing [5]. To form clusters among vehi- cles on the road, adaptive learning mechanisms have been proposed in the literature [5]. Computational intelligence techniques have been widely used in various applications in VANETs. The use of learning automata (LA) is one such technique that can be used in wide areas of applications such as call admission con- 1536-1284/14/$25.00 © 2014 IEEE Neeraj Kumar is with Thapar University Patiala (Punjab). Sudip Misra and Bibud- hendu Pati are with the Indian Institute of Tech- nology Kharagpur. Mohammad S. Obaidat is with Monmouth University. Joel J. P. C. Rodrigues is with Instituto de Teleco- municações, University of Beira Interior and Univer- sity ITMO. The work of J. Rodrigues has been par- tially supported by Insti- tuto de Telecomunicações, Next Generation Networks and Applications Group (NetGNA), by the Gov- ernment of the Russian Federation, Grant 074- U01, and by National Funding from the FCT — Fundação para a Ciência e Tecnologia through the Pest- OE/EEI/LA0008/2013 Project. M OBILE C ONVERGED N ETWORKS NEERAJ KUMAR, SUDIP MISRA, MOHAMMAD S. OBAIDAT , JOEL J. P. C. RODRIGUES, AND BIBUDHENDU P ATI ABSTRACT Due to the stringent constraints of constant topological changes and low end-to-end delay, data forwarding in the vehicular enviornment is always a challenging task. In this article, we have analyzed the performance of networks of learn- ing automata (LA) using the concepts of the Bayesian coalition game in the vehicular environ- ment. LA are assumed to be the players in the game, which form a coalition based on some pre- defined strategy from the strategy space. Each action taken by the players in the game may be rewarded or penalized by the environment in which they operate. The environment provides a feedback for each action taken by the LA. The probability of selection of an action is estimated using the Bayesian conditional probability on the payoff corresponding to each player. After fetch- ing input from the environment, the LA update their action probability vector. The performance of the proposed scheme is evaluated in different network conditions in a real environment by varying the learning rates of the automaton. A 20–30 percent enhancement of the successful packet delivery ratio has been observed using the proposed scheme in the vehicular environment. N ETWORKS OF L EARNING A UTOMATA FOR THE V EHICULAR E NVIRONMENT : A P ERFORMANCE A NALYSIS S TUDY