Workshop "From Objects to Agents" (WOA 2019) A Preliminary Study for an Agent Blockchain-Based Framework Supporting Dynamic Car-Pooling Maria Nadia Postorino∗ , Giuseppe M. L. Sarné‡ ∗ Department DICAM, University of Bologna, Viale Risorgimento 2, 40136 Bologna, Italy ‡ Department DICEAM, University of Reggio Calabria, Loc. Feo di Vito, 89122 Reggio Cal., Italy, sarne@unirc.it Abstract—In the last decades, private cars caused an increasing alternative to their ownership. The SE has gained relevance growth of urban traffic flows all over the world with a consequent in recent years thanks to i) advancements in technological increase of environmental pollution and road congestion. In areas (e.g., computer science, communication, electronics and this context, Car-Pooling is an alternative car-based solution for private mobility that optimizes the car loading factor with so on), which realize new ways to match demand and offer in respect to the number of passengers, although it requires that real time [20] and ii) cultural changes in people’s habits [21] all the participants share trip origin and destination at the same included an increasing environmental awareness. time. To make the system more appealing, an on-demand service In this respect, a general consensus there is on CP as adopting variable fares on the basis of trip length and number an efficient car-based solution for private mobility, which of participants is proposed in this paper. Multi-agent, reputation and blockchain technologies are used and a suitable dynamic may optimize the car use with respect to the number of routing algorithm has been developed. Experiments on simulated passengers [22]. Unfortunately, the common CP model lacks data prove the potentiality of this approach. flexibility because it is generally applied among privates and Index Terms—Car-Pooling, Multi-Agent System, Reputation for trips made on a regular basis. In other words, it requires System. that participants share the same time and the same origin and destination for the outward and (usually) for the return jour- I. I NTRODUCTION neys. For such a reason, CP participants generally share also Nowadays, most part of urban traffic flows all over the other characteristics like belonging to the same community, world is due to private cars. In the last decades, they caused working in the same place, living in the same neighboring and increasing costs for congestion, fuel consumption, travel time so on. Currently, some Web-platforms mediate among demand and health among the others [1]–[6], in a context where the and offer to reach an audience as larger as possible at low majority of the vehicles move only their drivers. costs [19]. Such platforms enlarged the basis of potential CP To reduce the use of private cars, a wide range of strategies users and have reached a fair popularity (also thanks to addi- has been developed by local, national and overnational (e.g., tional services like reputation systems, insurance, clear fares the European Union) authorities. Most part of the actions and so on) [23], [24] but have a low flexibility with respect to adopted inside urban areas rely on policies for (i) discouraging the trip origin and/or destination of the potential participants. the use of private cars by introducing restrictions as well as Indeed, their main aim is to facilitate an agreement among increasing their operational costs [7]–[13] and (ii) improving unknown participants for a CP journey. and promoting transit [14]–[16]. However, people continue to To give more flexibility in time and space to the traditional perceive private cars as more comfortable and flexible than CP, dynamic issues have been introduced recently [25], [26], transit thanks to the absence of constraints on time and space which try to match in real-time the mobility demand with the (e.g., fixed stops and timetables) that, conversely, characterize vehicles potentially able to satisfy it on-the-fly, if the new transit [17]. destination is compatible with those of the other passengers In the scenario above described, alternative forms of mobil- within prefixed time thresholds. Obviously, the other passen- ity still based on the use of cars are car-sharing (CS) and car- gers accept a priori to adapt their route for a lower trip fare. pooling (CP) systems. The CS is a Pay-As-You-Use modality In other words, when a ridesharing request is compatible with mainly adopted for short trips in urban areas, while in the those of the other passengers, then the road path is modified to CP two or more users that make frequently the same journey, pick and drop the new passenger, like a shared taxi. Obviously, at compatible times, voluntarily agree to share a private car demand and offer must be processed in an automatic way as and the traveling costs. The business models of these forms of quickly as possible to provide a real-time service. mobility belong to the “Sharing Economy” (SE) [18], [19], This study wants to contribute to this issue by proposing which promotes the temporary acquisition and shared usage a framework based on software agents, reputation system and of goods, services or knowledge as a possible and effective blockchain [27] technologies. More in detail, each user is sup- ported by a personal agent associated with his/her smartphone Maria Nadia Postorino, previously at Department DICEAM, University of Reggio Calabria, Loc. Feo di Vito, 89122 Reggio Cal., Italy, npos- (with suitable computation, storage, communication and GPS torino@unirc.it capabilities). These agents allow to identify and localize their 65 Workshop "From Objects to Agents" (WOA 2019) users by exploiting the smartphone GPS. This information is active on F at a certain time (see below); ii) manages and periodically sent to the Agency that, by means of a dynamic updates a reputation system based on the users’ feedback routing algorithm, matches demand and offer for rides by about the travel mates; iii) collects all the P As requests taking into account, positions, seat availability and temporal for rides. The Ag uses these information to search the constraints in terms of both departure and arrival times of all best opportunities for rides that, suitably ordered, are sent the participants. to the P As requiring them. Note that communication For each ridesharing request coming from a personal agent, among P As should occur via Ag by using a suitable the Agency will return a list of the existing ride opportunities tool. When a ride request/offer is accepted by the parts to that agent. From this list, the personal agent will select then commitment and payments (included compensations those rides considered as the most suitable for its user. Note due to the dynamic nature of this CP service) are carried that over time user’s choices provide information about his/her out via a smart-contract running on the permissioned preferences to its agent, which could allow a higher level blockchain platform. of customization to be realized. After a ride is selected, Similarly, the tasks carried out by a P A are: and confirmed by the driver, a smart-contract [28] runs over • Affiliation. In F each P A needs to be affiliated with Ag. a blockchain (like Ethereum [29]), to make the agreement • Activation. Whenever a P A enters (i.e., it is active) on public, irrevocable and for realizing safe payments in an F then, periodically, it sends its GPS coordinates to Ag. automatic way by using a cryptocurrency (e.g., Ether). • Search. To search a ride, a P A sends to Ag time and For each trip, it is also assumed a dynamic fare which de- position of both origin and destination point of its trip. pends both on how much long is the trip and how many riders Ag will answer with a list of opportunities, then the share that trip because the higher the number of participants P A will order this list, on the basis of the preferences to a ride is, the lower the comfort is as well as the cost for of its user, and will submit it to him/her for his/her passenger. It implies that in a dynamic CP, where the number choice (however, P A could perform this task also in an of passengers can change along the trip if a new rider is added, autonomous way [31]). the other participants receive a monetary compensation for • Commitment. After the choice of a ride, the P As of both both the loss of comfort and the decreased cost for passenger. driver and new passenger make public and irrevocable This procedure is automatically carried out each time that a their agreement by means of a smart-contract which runs smart-contract for that ride runs on the blockchain. on a permissioned blockchain and also provides monetary The paper is organized as follows. Section II presents payments and compensations with a cryptocurrency. the proposed agent framework, while Section III describes • Feedback. At the end of a ride each P A sends a feedback the main characteristics of the dynamic routing algorithm. f ∈ [0, 1] ∈ R (provided by its user) to Ag; the greater The reputation system is described in Section IV, while the f is, the greater its appreciation about the ride is. results of a campaign of simulations are show in Section V. Section VI gives an overview on the related work and, finally, Finally, the third component of F is a permissioned in Section VII some conclusions are presented. blockchain which allows to trust anonymous and unknown actors and warranties data integrity and payments without II. T HE M ULTI - AGENT F RAMEWORK exploiting other centralized third parties [32]. More in detail, each time that two users (i.e., a driver and a new potential This section describes a multi-agent framework (F ) able passenger) agree for a ride, then a smart-contract starts on to support a dynamic CP. The framework consists of three the adopted blockchain platform by using a permissioned components: approach. The smart-contract makes this choice irrevocable, • the Agency (Ag), which is a trusted and safe centralized publicly and realizes all the contractual obligations, payment component, unique in F , that coordinates the CP service and possible compensations to the included travel mates, in and collaborates with the other agents. an automatic way. To this aim, the information about the P A • A Personal Agent (P A), which supports the CP activities positions provided by Ag is also exploited. The costs due to of a user and runs on his/her personal device (i.e., smart- the blockchain management are assumed to be included in the phone) equipped with suitable computational, storing and CP fare. Note that this mechanism is not linked to a specific communication capabilities as well as with a pair of blockchain protocol1 . cryptographic keys [30] (for authentication and privacy issues). A P A is free of entering/leaving F at any time. III. T HE DYNAMIC ROUTING A LGORITHM • A permissioned blockchain. In this Section the algorithm for the dynamic route research Briefly, the tasks carried out by Ag are: running on the Agency is introduced. It is a variant of the • Affiliation. Ag manages the affiliation of each P A with Alpha-Beta Pruning algorithm [33] driven in the in-depth F and provides each P A with an identifier, an initial 1 For sake of simplicity, in this preliminary phase we refer to the well reputation score, and a pair of cryptographic keys. known Ethereum blockchain platform for the advantages deriving by both • Support. Ag manages its identity and all the CP service the availability of documented API with the opportunity of adopting its tasks. In detail, Ag: i) collects the position of all the P As cryptocurrency (i.e, Ether) and wallet service. 66 Workshop "From Objects to Agents" (WOA 2019) BF DFS M αP A∗ G DR Let N be a list of nodes Time complexity bd bd bm bm bd bd d ÷ bm Space complexity bd bd bm bm bd bd bm heuristic boundary nodes X X X T Exit by providing the complete heuristic search X X X Is N empty? solution (if it has been found), heuristic B&B X otherwise the partial one optimal solution X X X X X F complete search X X X X X partial solution X X X Split N in the subsets N1 and N2. N1 stores the nodes that do not generate cycles --ordered global path minimization X X according to the heuristic f1 (n) = d (n) + h ’(n) -- tree cuts X X X and N2 stores the nodes that generate cycles TABLE I Extract the first node n from the list N1, which becomes the new current state, C OMPARISON WITH B RUTE F ORCE (BF); D EPTH F IRST S EARCH and remove it from the list itself B RANCH &B OUND (DFS); M INIMUM (M); αP RUNING (αP); A+ ; DYNAMIC ROUTING (DR) - ( B - BRANCHING FACTOR , M - MAX DEPTH ). T Update the current best route Is n a goal node? if the new solution is better F search by suitable heuristics capable to improve the cut-off technique and to evaluate boundary nodes (if the chosen depth End the search on this branch Is horizon does not allow to reach the optimal solution). (pruning) f2 (n) = g (n) + h ’(n) > cost of the T current The main features of this algorithm (see Table I) are: best-line? • if the solution is contained into the fixed search depth Propagate to the parent node the F (horizon) then the optimal solution is provided, otherwise value of the border node, given by the heuristic the more promising one is provided. At each node, the f2 (n) = g (n) + h ’(n) Is the max depth of search algorithm optimizes a function (e.g., cost, number of T reached? passengers), positive and depending on one or more F Propagate to the parent parameters; node the best f2 (n) value received from its successors (from the • if the search ends in advance (e.g., for time or resource leaves towards the root) Is N empty? limits) the temporary optimal solution is provided; F • If cycles are allowed, the nodes generating them are T examined last, otherwise discharged if the solution cannot Fig. 1. The Dynamic Routing Algorithm in an iterative version. admit cycles; • when boundary conditions change, then the optimal route is recomputed; containing cycles, but these paths are examined last because • heuristic functions are adopted to estimate partial solu- the priority in expanding nodes is given to those nodes not tions, optimize the Branch & Bound and drive the in- present in the current path. Moreover, if it is impossible to depth search (the order of optimal expansion of nodes is find the optimal solution with respect the initial constraints chosen based on the closeness of nodes to the optimal (e.g., search time), the best solution found at the set depth solution); search is provided. Note that the fixed search depth does not • any list of expanded nodes for the Depth-First-Search affect the algorithm performance if it is greater than the depth (DFS) or for a temporary tree is stored. where the optimal solution is. In Figure 1, the pseudo-code (in form of flow-chart) of the IV. T HE R EPUTATION M ODEL algorithm is depicted. For sake of clarity, it is represented in an iterative form even if the algorithm is recursive. Note that The proposed reputation model considers the whole past the variables are appropriately initialized, the functions and history of each CP actor on the basis of the feedback assigned the adopted heuristic are known and the term “root” refers to to him/her by his/her counterparts [34]–[36]. Note that each the node from which the search starts. user has an initial reputation, set to 0.5, to contrast whitewash- When an intermediate node to be served is added, as in ing strategies [37] without penalizing too much the newcomers the dynamic CP, the algorithm starts a new search to find a with a low initial reputation value [38]. new path if the request is compatible with existing constraints In particular, when a ride ends, each user entrusts his/her (e.g., time and reputation), otherwise the request is rejected. appreciation in terms of feedback (f ∈ [0, 1] ⊂ R) to his/her The algorithm solves a route problem linking more nodes P A that, in turn, sends f to Ag which manages the reputation by considering (without loss of generality) the problem of system. When a feedback fx,y , released by the user ux about searching (on the basis of the adopted heuristic) the best path the user uy , is received h by the Ag,  this latteri computes the between two consecutive nodes as independent from all the parameter Qx,y = b Px,y + rx,y · ax · fx,y /2, where: i) other paths linking the other nodes belonging to the route P is the final cost of the ride; ii) rx,y is the number of past (i.e., local solution). Even though in the CP problem the rides between ux and uy ; iii) ax is the accuracy of ux in cycles should be avoided, the algorithm can also find solutions providing a feedback; iv) fx,y is the feedback given by ux 67 Workshop "From Objects to Agents" (WOA 2019) about uy . To hinder malicious behaviors, Ag will update the 1.0 case A current reputation score of uy (i.e., Ryold ) as Rynew = α · case B % identified users 0.8 Ryold + (1 − α) · Qj,i only if Qj,i > 0 ∨ Ryold ≥ 0.5 is true, 0.6 conversely Ry will not be updated (e.g., Rynew = Ryold ). The parameter α ∈ [0, 1] ⊂ R gives more or less relevance to Ryold 0.4 with respect to the new contribution Qx,y . 0.2 To avoid alternate behaviors, the parameter Pb takes into account the cost P of the ride, where PM ax is the maxi- 0 0 5 10 15 20 25 mum cost threshold, such that Pb = M in(1, P/PM ax ). The epochs parameter r is effective against collusive behaviors occurring among two or more users that frequently exchange positive Fig. 2. Percentage of users correctly identified. Cases A and B. feedback to increase maliciously their reputation scores. To this aim, let Tx,y be a parameter depending on the time occurring between two consecutive interactions between ux the users’ nature very quickly (e.g., 90% of users is correctly and uy evaluated positively, the variable rx,y is computed as classified in less than 11 and 18 epochs for the cases A  and B, respectively). With respect to similar systems, this is rx,y = 1 (e(1−Tx,y ) ) if fx,y ≥ 0.5 ∧ Tx,y > 1 , otherwise rx,y = 1. To compute Tx,y , let tl and tp be the timestamps of also due to the fact that each user receives more feedback the last two positive feedback and let ∆T be a time threshold. for the same ride, depending on the (random) number of At the first positive feedback Tx,y = 1 while, for each further participants to the ride. Note that, all the “neutral” users’ positive feedback, i) if (|tl − tp | < ∆t then Tx,y = Tx,y + 1, behavior, receiving an initial reputation set to 0.5, are correctly otherwise ii) Tx,y = M ax[1, Tx,y − ⌊|tl − tp |/∆t⌋]. recognized from the first to the last epoch. However, also by The accuracy degree of ux in providing a correct feedback is setting a different value for the initial reputation, these users taken into account by the parameter ax ∈ [0, 1] ∈ R where 1/0 are quickly recognized like the other. stands for maximum/minimum accuracy. More specifically, ax The second experiment verified the performance of the is computed as axnew = β · axold + (1 − β) · 1 − |fx,y − Ry | , dynamic routing algorithm with different graphs (even large where the parameter beta ∈ [0, 1] ⊂ R weights aold with ones) generated randomly by adopting on the parameters com- x respect to the new contribution given by the difference between patible with the considered scenario. For each new admissible fx,y and the reputation of the target user (i.e., Ry ). request the route is modified by rejecting the requests incom- patible with both pre-existing time constraints and minimum V. E XPERIMENTS reputation and by choosing among: i) minimizing the cost; ii) maximizing the number of served users; ii) maximizing the This section presents the results of two experiments testing Users/Cost ratio. The computing time is always negligible, the reputation system and the dynamic routing algorithm. some milliseconds. From an operational point of view, note To test the reputation system, a population of 1000 users that when the new passengers are collected at a node, the randomly distributed on 5 behaviors (e.g., very unpleasant, un- requests on the previous node is reset to avoid the insertion pleasant, neutral, pleasant, very pleasant), has been simulated of loops between the two nodes. for unvarying (case A) and alternate2 (case B) modalities. In the simulations, the reputation system parameters α and VI. R ELATED W ORK β have been set both to 0.2 and Pmax has been set to In recent years, Intelligent Transport Systems (ITSs) re- 5e. The number of ride mates and cost of the ride have ceived a great impulse from advancements occurred in com- been randomly generated respectively in the ranges 2 ÷ 5 and [1.00e, 7.50e]. Moreover, in the case B, the cost of the CP services has been set > 5e for the 25% of the rides. The simulation has been arranged for epochs, so that the reputation parameters tl and tp have been set to the respective epochs and ∆t has been set to 3 epochs. Each feedback has been randomly generated coherently with the user’s behavior. All the users received an initial reputation score of 0.5 and for each epoch only a population share of 20% has been randomly selected. Obviously, the higher the percentage of users correctly identified, the higher the accuracy of this reputation system. Figure 2 shows the results obtained for the two cases A and B on the basis of the user’s behavior correctly recognized. A common aspect is represented by the ability to recognize 2 With respect to the ride cost. Fig. 3. Random network 29 Nodes, 51 Links, path from 13 to 7, depth 6. 68 Workshop "From Objects to Agents" (WOA 2019) sible. Since the blochchain is replicated on more independent hosts, it cannot be easily controlled, tampered or deleted [59]. Note that blockchain performance are significantly affected by the adopted consensus protocol in terms of computational complexity, robustness, latency, scalability and safety. Behind the cryptocurrencies (like Bitcoin [27]) some blockchains can realize smart-contracts [28], i.e. computerized transactions that realize the terms of a contract. For instance. Ethereum [29] was the first blockchain for smart-contracts but, nowadays, other similar platforms exist (e.g., Ripple [60], Stellar [61] and Tendermint [62]) and, like Ethereum, most of them has their own digital coin (like Ether). This class of blockchains appears the most suitable for developing the sharing economy Fig. 4. Random Network 500 Nodes, 306 Links, path from 119 to 361, depth business. 10. VII. C ONCLUSIONS puter science, electronic and communication above all. In this CP can contribute to support public and private mobility by context, an increasing attention is given to the adoption of reducing urban traffic and its environmental problems. To this the intelligent software agent technologies [39]–[42] thanks aim, in this paper a dynamic form of CP potentially able to to their learning and adaptive capabilities [43], the attitude to enlarge the number of CP users has been presented. cooperate by sharing their These issues have been addressed by exploiting multi-  knowledge [44]–[47] and to deal agent systems, reputation systems and blockchain technologies with large, uncertain and or dynamic systems in a centralized or distributed way. and tested by realizing some experiments on simulated data. The first experiment verified the capability of the proposed Also trust and reputation systems are adopted in the ITS algorithm to manage the dynamic routing, while the second area, mainly because in larger and larger dynamics environ- one, based on simulated data, verified the effectiveness of the ments information about counterparts is often necessary [48], proposed reputation system for two scenarios. The results of [49] to evaluate the trustworthiness of potential counterparts. these preliminary experiments encourage further developments Note that to assure integrity of both trust/reputations and of this form of dynamic CP. identities, also cryptographic techniques are often exploited. In the CP scenario, reputation systems are worthy to be ACKNOWLEDGMENT implemented [50], [51] and several models exist. For instance, in [52] a reputation system is used to refuse undesirable pas- This work has been developed at the Networks and Complex sengers to avoid having unpleasant rides. Similarly, the authors Systems (NeCS) Laboratory - Department of Engineering of [53] and [54] propose to implement reputation systems, Civil, Energy, Environment and Materials (DICEAM) - Uni- respectively named Smart Rider Seeker and SmartShare, that versity Mediterranea of Reggio Calabria. allow drivers and commuters to offer and request rides by also R EFERENCES permitting to reject potential participants. However, all these approaches do not describe any specific reputation model, dif- [1] M. N. Postorino and G. M. L. Sarnè, “Mobility forecast in an urban area through the use of neural networks,” in Applications of advanced ferently from our proposal. CS activities too can benefit from technologies in transportation engineering. ASCE, 1995, pp. 213–217. reputation information, as in [55] where agents assist users [2] C. A. M. Toledo, “Congestion indicators and congestion impacts: a in improving their driving behavior by means of individual study on the relevance of area-wide indicators,” Procedia-Social and Behavioral Sciences, vol. 16, pp. 781–791, 2011. reputation measures, also used to obtain both the access to [3] L. Chen and H. Yang, “Managing congestion and emissions in road CS services and personalized fares. Some experiments on real networks with tolls and rebates,” Transportation Research Part B: and simulated data show the potentiality of this approach. Methodological, vol. 46, no. 8, pp. 933–948, 2012. [4] E. Cascetta and M. N. Postorino, “Fixed point approaches to the Blockchain [27] and smart-contracts are giving new oppor- estimation of o/d matrices using traffic counts on congested networks,” tunities both to multi-agent systems [56] and to the mobility Transportation science, vol. 35, no. 2, pp. 134–147, 2001. ecosystem to act in sharing, insurances, payment activities [5] A. Spickermann, V. Grienitz, and A. Heiko, “Heading towards a multimodal city of the future?: Multi-stakeholder scenarios for urban and store publicly and permanently car profiles, maintenance, mobility,” Technological Forecasting and Social Change, vol. 89, pp. accident, transfer and other data [57]. A blockchain is a 201–221, 2014. decentralized, distributed ledger of interconnected data block [6] T. Fontes, S. Pereira, P. Fernandes, J. Bandeira, and M. Coelho, “How to combine different microsimulation tools to assess the environmental that once added, in a chronological way, are permanent and impacts of road traffic? lessons and directions,” Transportation Research unchangeable. Before adding a data block, it has to be vali- Part D: Transport and Environment, vol. 34, pp. 293–306, 2015. dated by a distributed consensus protocol [58] based on three [7] M. N. Postorino, G. Musolino, and P. Velonà, “Evaluation of o/d trip matrices by traffic counts in transit systems,” in Schedule-Based steps (e.g., transaction endorsement, ordering, validation and Dynamic Transit Modeling: theory and applications. Springer, 2004, commitment) after which it will be added and publicly acces- pp. 197–216. 69 Workshop "From Objects to Agents" (WOA 2019) [8] S. Ison and T. Rye, “Implementing road user charging: the lessons learnt [36] M. N. Postorino and G. M. L. Sarnè, “A neural network to identify from hong kong, cambridge and central london,” Transport Reviews, driving habits and compute car-sharing users reputation,” in Italian vol. 25, no. 4, pp. 451–465, 2005. Workshop on Neural Nets. Springer, 2017, pp. 207–216. [9] N. Paulley, R. Balcombe, R. Mackett, H. Titheridge, J. Preston, M. Ward- [37] G. Zacharia and P. Maes, “Trust management through reputation mecha- man, J. Shires, and P. White, “The demand for public transport: The nisms,” Applied Artificial Intelligence, vol. 14, no. 9, pp. 881–907, 2000. effects of fares, quality of service, income and car ownership,” Transport [38] S. Ramchurn, D. Huynh, and N. Jennings, “Trust in multi-agent sys- Policy, vol. 13, no. 4, pp. 295–306, 2006. tems,” Knowledge Engeenering Review, vol. 19, no. 1, pp. 1–25, 2004. [10] S. Liu, K. P. Triantis, and S. Sarangi, “A framework for evaluating [39] B. Chen and H. Cheng, “A review of the applications of agent technology the dynamic impacts of a congestion pricing policy for a transportation in traffic and transportation systems,” Intelligent Transportation Systems, socioeconomic system,” Transportation Research Part A: Policy and IEEE Trans. on, vol. 11, no. 2, pp. 485–497, 2010. Practice, vol. 44, no. 8, pp. 596–608, 2010. [40] C. Adam and B. Gaudou, “Bdi agents in social simulations: a survey,” [11] M. N. Postorino, “A comparative analysis of different specifications of 2016. modal choice models in an urban area,” European journal of operational [41] J. P. Müller and K. Fischer, “Application impact of multi-agent systems research, vol. 71, no. 2, pp. 288–302, 1993. and technologies: a survey,” in Agent-Oriented Software Engineering. [12] A. de Palma and R. Lindsey, “Traffic congestion pricing methodologies Springer, 2014, pp. 27–53. and technologies,” Transportation Research Part C: Emerging Technolo- [42] M. N. Postorino and G. M. L. Sarné, “Agents meet traffic simulation, gies, vol. 19, no. 6, pp. 1377–1399, 2011. control and management: A review of selected recent contributions,” in [13] M. Gibson and M. Carnovale, “The effects of road pricing on driver Proc. of the 17th Workshop from Objects to Agents, WOA 2016, ser. behavior and air pollution,” J. of Urban Economics, vol. 89, pp. 62–73, CEUR Workshop Proceedings, vol. 1664. CEUR-WS.org, 2016. 2015. [43] L. Busoniu, R. Babuska, and B. De Schutter, “A comprehensive survey [14] J. D. Harford, “Congestion, pollution, and benefit-to-cost ratios of us of multiagent reinforcement learning,” Systems, Man, and Cybernetics public transit systems,” Transportation Research Part D: Transport and (C): Appl. and Reviews, IEEE Trans., vol. 38, no. 2, pp. 156–172, 2008. Environment, vol. 11, no. 1, pp. 45–58, 2006. [44] V. Tomás and L. Garcia, “A cooperative multiagent system for traffic [15] M. N. Postorino and V. Fedele, “The analytic hierarchy process to management and control,” in Proc. of the 4th Int. Joint Conf. on evaluate the quality of service in transit systems,” WIT Transactions Autonomous agents and multiagent systems. ACM, 2005, pp. 52–59. on The Built Environment, vol. 89, 2006. [45] F. Wang, “Agent-based control for networked traffic management sys- [16] C. Winston and V. Maheshri, “On the social desirability of urban rail tems,” Intelligent Systems, IEEE, vol. 20, no. 5, pp. 92–96, 2005. transit systems,” Journal of urban economics, vol. 62, no. 2, pp. 362– [46] B. Chen, H. Cheng, and J. Palen, “Integrating mobile agent technology 382, 2007. with multi-agent systems for distributed traffic detection and manage- [17] B. Caulfield, “An examination of the factors that impact upon multiple ment systems,” Transportation Research Part C: Emerging Technologies, vehicle ownership: The case of dublin, ireland,” Transport Policy, vol. 17, no. 1, pp. 1–10, 2009. vol. 19, no. 1, pp. 132–138, 2012. [47] S. Sen, A. Biswas, and S. Debnath, “Believing others: Pros and cons,” [18] J. Parsons, “Remix: Making art and commerce thrive in the hybrid in Proc. of the 4th Int. Conf. on Multi-Agent Systems, ICMAS’2000. economy,” Journal of Teaching and Learning, vol. 7, no. 1, 2010. IEEE, 2000, pp. 279–286. [19] R. Belk, “You are what you can access: Sharing and collaborative [48] D. Rosaci, G. M. L. Sarné, and S. Garruzzo, “Integrating trust measures consumption online,” J. of Business Research, vol. 67, no. 8, pp. 1595– in multiagent systems,” International Journal of Intelligent Systems, 1600, 2014. vol. 27, no. 1, pp. 1–15, 2012. [20] K. Dervojeda, “Accessibility based business models for peer-to-peer [49] M. N. Postorino and G. M. L. Sarné, “An agent-based sensor grid to markets,” Business Innovation Observatory: The Sharing Economy, Case monitor urban traffic,” in Proc. of the 15th Workshop “from Objects study, vol. 12, 2013. to Agents”, WOA 2014, ser. CEUR Workshop Proceedings, vol. 1260. [21] J. Agyeman, D. McLaren, and A. Schaefer-Borrego, “Sharing cities,” CEUR-WS.org, 2014. Friends of the Earth Briefing, pp. 1–32, 2013. [50] D. Graziotin, “An analysis of issues against the adoption of dynamic [22] B. Cohen and J. Kietzmann, “Ride on! mobility business models for carpooling,” arXiv preprint arXiv:1306.0361, 2013. the sharing economy,” Organization & Environment, vol. 27, no. 3, pp. [51] C. Caballero-Gil, P. Caballero-Gil, J. Molina-Gil, F. Martı́n-Fernández, 279–296, 2014. and V. Loia, “Trust-based cooperative social system applied to a carpool- [23] https://www.blablacar.it, 2019. ing platform for smartphones,” Sensors, vol. 17, no. 2, p. 245, 2017. [24] https://www.singucarpooling.com, 2019. [52] D. B. Nagare, K. L. More, N. S. Tanwar, S. Kulkarni, and K. C. Gunda, [25] J. Friginal, S. Gambs, J. Guiochet, and M.-O. Killijian, “Towards “Dynamic carpooling application development on android platform,” privacy-driven design of a dynamic carpooling system,” Pervasive and International Journal of Innovative Technology and Exploring Engineer- mobile computing, vol. 14, pp. 71–82, 2014. ing, vol. 2, no. 3, pp. 136–139, 2013. [26] J. Alonso-Mora, S. Samaranayake, A. Wallar, E. Frazzoli, and D. Rus, [53] S. Abdel-Naby and P. Giorgini, “Smart ride seeker (srs) an introductory “On-demand high-capacity ride-sharing via dynamic trip-vehicle assign- plan,” University of Trento, Tech. Rep., 2006. ment,” Proc. of National Academy of Sciences, vol. 114, no. 3, pp. 462– [54] H. Packer and L. Moreau, “A methodology to take account of diversity 467, 2017. in collective adaptive system,” 2016. [27] S. Nakamoto, “Bitcoin: A peer-to-peer electronic cash system,” 2008. [55] E. Picasso, M. N. Postorino, and G. M. L. Sarné, “A study to promote [28] N. Szabo, “Smart contracts,” Unpublished manuscript, 1994. car-sharing by adopting a reputation system in a multi-agent context.” [29] https://www.ethereum.org, 2019. in Proc. of the 18th Workshop from Objects to Agents, WOA 2017, ser. [30] C. Adams and S. Lloyd, Understanding PKI: concepts, standards, and CEUR Workshop Proceedings, vol. 1867. CEUR-WS.org, 2017, pp. deployment considerations. Addison-Wesley Professional, 2003. 13–18. [31] M. N. Postorino and G. M. L. Sarne, “A neural network hybrid [56] G. Ciatto, A. Maffi, S. Mariani, and A. Omicini, “Towards agent-oriented recommender system,” in Proceedings of the 2011 conference on neural blockchains: Autonomous smart contracts,” in PAAMS 2019, 2019. Nets WIRN10, 2011, pp. 180–187. [57] D. Namiot, O. Pokusaev, V. Kupriyanovsky, and A. Akimov, [32] M. A. Khan and K. Salah, “Iot security: Review, blockchain solutions, “Blockchain applications for transport industry,” International Journal and open challenges,” Future Generation Computer Systems, vol. 82, of Open Information Technologies, vol. 5, no. 12, pp. 130–134, 2017. pp. 395–411, 2018. [58] N. Chalaemwongwan and W. Kurutach, “State of the art and challenges [33] D. E. Knuth and R. W. Moore, “An analysis of alpha-beta pruning,” facing consensus protocols on blockchain,” in Information Networking Artificial intelligence, vol. 6, no. 4, pp. 293–326, 1975. (ICOIN), 2018 International Conference on. IEEE, 2018, pp. 957–962. [34] P. De Meo, F. Messina, M. N. Postorino, D. Rosaci, and G. M. L. Sarné, [59] M. Pilkington, “11 blockchain technology: principles and applications,” “A reputation framework to share resources into iot-based environ- Research handbook on digital transformations, p. 225, 2016. ments,” in 2017 IEEE 14th International Conference on Networking, [60] https://ripple.com/, 2019. Sensing and Control (ICNSC). IEEE, 2017, pp. 513–518. [61] https://www.stellar.org, 2019. [35] P. De Meo, F. Messina, D. Rosaci, and G. M. L. Sarné, “Combining [62] https://tendermint.com, 2019. trust and skills evaluation to form e-learning classes in online social networks,” Information Sciences, vol. 405, pp. 107–122, 2017. 70