Creating a Social Brain for Cooperative Connected Autonomous Vehicles: Issues and Challenges

10/02/2017 ∙ by Seng W. Loke, et al. ∙ Deakin University 0

The connected autonomous vehicle has been often touted as a technology that will become pervasive in society in the near future. Rather than being stand alone, we examine the need for autonomous vehicles to cooperate and interact within their socio-cyber-physical environments, including the problems cooperation will solve, but also the issues and challenges.



There are no comments yet.


page 5

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Connected autonomous vehicles (CAVs) are predicted to revolutionise transportation world-wide, and transform urban life, as early as 2021, and becoming pervasive in the decades to come.111See,, around the world testing of autonomous vehicles (see, NHTSA guidelines on development of automated vehicles:, and self-driving vehicle initiative in Australia (

Much work has been on sensors for vehicles and how the Artificial Intelligence (AI) in vehicles can learn to “ see” , navigate and manoeuvre within everyday road systems. In the larger Internet-of-Things (IoT) service ecosystems, CAVs are situated within sociotechnical environments of human road users and other autonomous entities. In so far as the action of CAVs must take into account the actions and reactions of others, and are intentional individually or collectively, CAVs perform social actions.

Vehicles need to interact with and potentially connect not just to other vehicles, but also motorcycles, bicycles, pedestrians, and other road-users, as well as with IoT services (including via Road-Side-Units), over Dedicated Short Range Networking (DSRC) or 5G-V2X networking. Indeed, there have been much research on vehicle-to-vehicle (v2v) and vehicle-to-infrastructure (v2i) (and more generally, v2x) communications. Over such network protocols, there is a vast opportunity for vehicles to exchange application level messages and cooperate to improve safety and increase their effectiveness, creating a cooperation layer above the vehicular network layer. The Society of Automotive Engineers (SAE) released a message set dictionary for standardizing messages exchanged in DSRC communications, such as intersection collision warnings, emergency vehicle alerts and vehicle status information can be shared.222

CAVs will not just get from A to B autonomously but cooperate autonomously in doing so. How vehicles should autonomously cooperate with other vehicles, pedestrians and services using such v2x networking, how vehicles can reason about social behaviours, how vehicles should behave when receiving particular messages, how inter-vehicular cooperation can be exploited, and how CAVs should interact with people, are still emerging areas of research - that is, it is not yet clear what the social brain of CAVs need to be.

Work on the social brain of CAVs is just emerging. At the same time, policy rules and robot laws are needed to govern not just road traffic in general but to ensure the trustworthy and ethical interaction between CAVs, and between CAVs and people [17].

This paper aims to outline the cooperation layer in CAVs, highlighting potential applications and issues of cooperative CAVs, drawing on related work. We first describe a range of scenarios where cooperation can happen as well as the potential benefits, and then outline challenges in enabling the social brain of CAVs.

2 Types of Cooperation

We consider below three types of cooperation: vehicle-to-vehicle, vehicle-to-thing, and vehicle-to-pedestrian.

2.1 Vehicle-to-Vehicle Cooperation

We consider a range of scenarios, including parking, routing, managing traffic flows, and social memory, to illustrate the potential that v2v cooperation in addressing important on-road problems.

Parking. Cooperation is not merely sharing of information. For example, the use of inter-vehicle cooperation in finding parking spaces have been investigated in different aspects. The authors in [12] analysed the performance of cooperative vehicles in finding parking spaces through collecting and sharing parking information compared with centralized management under different conditions and concluded that there is no optimal solution for all situations and the benefit of sharing information can outweigh the increased vehicle competition that it may cause. Also, the authors in [22] found through simulations that disseminating parking information among vehicles barely decreased the searching time and even occasionally increased the searching time. The work in [5] aimed to reduce contention among vehicles by restricting sharing to only information that the vehicle itself is not interested in. In [2] is discussed a decentralised car parking allocation mechanism inside a car park by supporting the vehicles with initial information about available slots at the car park gate and using vehicular cooperation that shares intentions about where to park and negotiates to resolve competition - a reduction of time-to-park of up to 25% is possible. Hence, it is not necessarily just sharing information, but cooperation via negotiation and coordination. There are other notions of interaction among cars when parking (e.g., a car leaves a note for another saying “let me know when you leave” etc).

Routing. Recent work in vehicle-to-vehicle cooperation explored vehicles coordinating their routes, and in doing so, can distribute themselves along faster, even if longer, routes. In [8], by having cars opportunistically cooperating on routes, over DSRC v2v communication, when they come near each other at disparate intersections, traffic congestion across a vast area can be alleviated; vehicles can get to destinations sooner by as much as 30% compared to all just taking the shortest distance route. It is noteworthy that some gains can be obtained without global cooperation among all vehicles, but merely by local cooperation within groups of vehicles at intersections. This suggests some cooperation even locally where possible can have far reaching consequences. The work in [16] showed that inter-vehicle communication can help to detect congestion and to suggest new routes to avoid congestion.

Swarm Behaviours for Dynamic Traffic Flows. Traditionally, markings on the road or traffic signals are used to coordinate vehicles so that they move in an orderly manner, but they can cause delays or reduce the utility of roads. With cooperation and cyber-coordination, traffic flow can be coordinated without physical signals. MIT has been working on intersections without traffic lights based on a slot-based system (based on Intersection Managers).333

Also, cooperation among vehicles can also be used to form flexible collective vehicle behaviours. These possibilities are illustrated in Figure 1. For example, on a highway, lanes are fixed equally on both sides (e.g., five lanes in each direction) - while at certain times, traffic can be much heavier in one direction than the other. With CAVs, when traffic volume is high in one direction, vehicles could cooperate massively and inform other vehicles that there are eight lanes now in one direction and two in the reverse way. Lanes can then be re-balanced at other times. Also, some “lanes” can become narrower (with cars moving closer to each other but slower) at certain times, while at other times lanes are broader (with cars moving faster but further apart from each other). Another scenario is making way for an emergency vehicle passing through - CAVs that receive notifications of the emergency vehicle can coordinate their actions to create a path way. All vehicles in the vicinity could receive the same message from the emergency vehicle, but each need to decide what is the best action to take so that a path is created, based on contextual knowledge of itself and its surrounding vehicles. The vehicles must also be able to detect that the emergency vehicle has passed through so that they can cooperatively resume their normal movements.

Figure 1: Swarm-Based Dynamic Traffic Flow

Platooning and Safety. There are many forms of v2v cooperation on the move. For example, vehicles can opportunistically platoon to improve travel times, road usage, and safely do so [4]. Vehicle platooning on highways has been widely explored [10]. Cooperative adaptive cruise control is already explored [18], where movement models of vehicles are shared, via v2v communications, rather than raw data about locations.

Collision warning at intersections and at hard-to-see settings via v2v communication is widely noted.444E.g., see Also, vehicle to bicycle and bicycle to bicycle interactions can enhance safety [9].

Cooperation between vehicles entering a highway and vehicles on the highway can enable safer traffic merging, and can be achieved via vehicle intention recognition based on speeds [24] and v2v communication [25].

Long Term Cooperation - Social Networks and Social Memory.

Future cars could have social memory. Cars can record past pattern of interactions and connections with other cars and things, forming conceptual links with objects in the world, e.g., labelling some links as “car-friend”. A car can remember the favours received by another car, and so, reciprocate when it gets a chance.

Such remembered links and relationships do not mean that cars cannot communicate with one another as needed or ad hoc (in the same that people have social networked friends but also can interact with anyone who are not in their social network, and add social network friends). The advantage of such remembered connections and relationships is to track regularity of pattern of interaction and perhaps even establish a trust network for information exchange, reciprocal (favour exchange) interactions, and to perform problem-solving over such networks. For example, cars take turns to give way to one another or defer parking spots to one another at different times. The social memory provides a context for future v2v communications, yielding greater efficiency in expressing intent.

2.2 Vehicle-to-Thing Cooperation

Recent work has explored the use of drones coupled with vehicles, e.g., where vehicles are drone stations, from which drones to survey disaster-struck areas are deployed. Drones can be used to guide self-driving vehicles or do delivery on behalf of such vehicles,555As an example, see the DJI-Ford Challenge ( and the recent Ford’s patent on self-driving vehicles with detachable drone ( or a collection of vehicles can band together to send a drone out to look ahead to observe in detail the situation far ahead, or around the corner.

There are also other possibilities such as wheelchairs being integrated with self-driving vehicles, so that an integrated system of door-to-door transportation can be made - e.g., wheelchair coordinating with self-driving vehicles on where to drop-off and pick-up. Self-driving vehicles can also coordinate with not only people, but robots and things, for their pick-up or drop-off. A thing or robot might require repairs and so crowdsources self-driving vehicles to send it to a repairer - all automated, if the owner pre-authorizes. An autonomous vehicle may be sent to pick up a broken down autonomous vehicle.

Autonomous vehicles can also coordinate with fueling (or charging) stations automatically, in determining appropriate schedules in order to reduce wait times.

A road device could provide information about traffic and situation-awareness. A car passing by it could exploit such information for safety (speed), comfort, mobility (traffic), emissions (speed), or decision-making.

2.3 Vehicle-to-Pedestrian Cooperation

CAVs will share roads with pedestrians, so that the interaction between a vehicle and the humans outside the vehicle is an issue to address [1, 20].666, Human-vehicle interaction mediated by v2x communication (e.g., vehicle to wearable device) can provide a greater level of communication between human and vehicle, complementary to, or as an alternative to, visual physical gestures or signaling.

3 Challenges

3.1 How Should Vehicles Talk to Each Other?

While standardized message formats will help to enable interoperability, the way in which vehicles should or must respond to a given message remains difficult to specify. An interesting direction is to consider how multiagent communication languages [13] developed for specifying meaningful interaction between software agents, based on speech act theory777 can be reinterpreted in the context of v2x cooperation:

  • assertives: sharing how things are, when cars share status or environmental information, e.g., parking information, observations of road situations, and so on,

  • directives: orders and commands could be issued from authoritative vehicles or persons to an autonomous vehicle, calling for certain behaviours, e.g., a traffic police officer commanding an AV to stop or an emergency vehicle commanding vehicles to make way - AVs, when receiving such directives can authenticate the senders to ensure that the directives are from authoritative parties before complying,

  • commissives: these include promises and vows; pedestrians or vehicles may need to act based on the commitment of other vehicles or pedestrians - e.g., when a vehicle promises to give way to another vehicle, a commissive message is sent, and it should behave accordingly, but if it does not, it can be held accountable for not doing so (a blackbox system that records such commissives can be analyzed against actual behaviour),

  • expressives: expressive feelings and attitudes (e.g., apologies and thanks) seems out of place for robot cars, but might be useful when interacting with humans, e.g., an AV thanks a human driver for giving way,

  • declarations: these are messages that give rise to social world structures and institutions, and might represent collective intentionality in the sense of Searle [21] - an example with reference to the above dynamic traffic flow scenario is where a collection of vehicles declares that the road is now one way for a period, and are allowed to do so in certain circumstances. Another example is vehicles coordinating their routes to go in certain directions. Collective recognition and acceptance of this decision effectively turns a two way road into a one way road, at least for that period. How such collective intentionalities cause behaviours constituting cooperation can be further explored.

3.2 Lawful Interactions

Often, the vehicles will need to follow traffic regulations and laws concerning what to do on receiving incident messages. For example, a recent road rule in Melbourne, Australia, is that vehicles must slow to no more than 40km/h when passing crashes and incidents where emergency vehicles are stopped on the side of the road. Hence, CAVs will need to be able to respond to that. In many countries, vehicles must make way for a government vehicle or an emergency vehicle passing through. It is not clear CAVs need to be programmed to be context/situation-aware and still respond safely to receiving messages. There has been recent attention on robot laws and building ethical or lawful behaviour into AI [23], though not focused on CAVs.

Situational rules might need to be represented within such vehicles, e.g., the fact that certain vehicles have certain rights (such as right of way) or certain obligations in particular contexts [21]: (in certain physical world positions) counts as (special car status with certain rights) in (some real world conditions).

There has been much work on norms and electronic institutions from multiagent research (e.g., [19]), with interesting applications to creating e-institutions for CAVs in order to regular CAV behaviours.

3.3 Scales of Cooperation

It is unlikely or unnecessary that all autonomous vehicles in a country or a city cooperate. Often, autonomous vehicles close to each other (e.g., a platoon) or within a locality (near an intersection, a car park or a stretch of a road) might cooperate. It can be shown that such local cooperation can lead to global effects - for example, cars coordinating routes when they meet at intersections is adequate to reduce traffic congestion on a much larger scale [8]. Also, a set of cars cooperating to park more efficiently in a part of the CBD could reduce traffic congestion for cars in other parts of the CBD. While there could be beneficial emergent effects on much larger scales, how one could engineer wide-scale benefits from disparate sets of vehicles cooperating locally is still an open research question.

3.4 Context-Aware Decision-Making and Regulations

The ability to understand the situation in particular regions is needed by CAVs in decision making. Consider the following example first introduced in [15], but simplified here. A self-driving (or autonomous) smart car can be programmed with a particular destination, and it could bring the passengers there, but upon arrival, there are a number of possibilities, two of which are:

  1. the autonomous car drops off the passengers (including the driver) and then either proceeds to find a car park nearby, or simply cruises around nearby; or

  2. the autonomous car allows the driver to take back control on nearing the destination.

Note that the second option would enable the driver to take over and it would proceed as a traditional car with the driver making decisions about what to do from then on, but takes away full autonomy. The first option is what we would expect but suppose the place turns out to have a huge traffic jam so that even simply dropping off the passengers would not be easy and the car might be stuck in traffic waiting for its turn at the drop-off zone. So, the car could choose try to drop the passengers off a bit further away from the main drop-off zone, with approval from the driver. Hence, the car has to know when it needs to involve its passengers in such decision-making, even if it is assumed that the passengers are simply leaving it to the car to take them to the right place.

3.5 Trusted Communications and Deception-Proofing

It is also not clear when and which messages should be trusted. Vehicles can receive incident messages which are false and react in ways that endanger other vehicles. Vehicles need to take into account their own situation/context when responding to other messages. There has been recent work on attempting to detect and deal with false or deceptive messages and verifying and validating messages in v2v communications [14], though not in the context of cooperation. Cooperative mechanisms should be relatively robust and resilient to invalid messages or deception, or even the misinterpretation of particular vehicles of a received message.

Also, vehicles can receive uncertain, false or malicious messages affecting their behaviour, or vehicles can form coalitions that can preclude non-coalition vehicles from taking advantage of cooperation. Hence, cooperation will also need to be robust against false messages or coalitional behaviours.

Recent research has begun to address this [7]

which showed that the probability of correct message delivery reduces to its minimum after the proportion of malicious vehicles in the network crosses a threshold, that is, trust requires a collective effort.

3.6 Standards: Cooperation Protocols and Behaviour

There may be a need for standards to define the language of messages that the CAVs can understand as well as the operational meaning of the messages. The operational meaning refers to how CAVs are to behave in response to such messages - should how CAVs respond, and not just the messages, be standardized? For example, when receiving an alert about emergency messages, the behaviour of the CAVs must be to carefully give way. There are questions about how the cooperation of vehicles given such emergency messages, even if they are from different manufacturers, can happen. For example, if each vehicle seeks to get out of the way safely in a certain pre-programmed, then will a clear way emerge for the emergency vehicle to pass through?

There are also issues with human-vehicle interaction - what protocols are needed for human drivers to interact with autonomous vehicles, and for pedestrians to interact with autonomous vehicles, in a standard way (for vehicles from all manufacturers). By having clear expectation of behaviours of CAVs in different road situations, humans might then be better able to adapt their behaviour around CAVs.

There could be a range of cooperation protocols for CAVs in different situations, e.g., a platooning protocol, a protocol for interaction at intersections, a protocol for directional swarming, a protocol for interaction in a car park, a protocol for traffic merging [3] and so on. High-level representations of CAV cooperative behaviour using multiagent models could be explored and protocols formally verified, e.g., as done for platooning protocols using a Belief-Desire-Intention agent model of vehicles [11], and using the pi-calculus to model v2v communication protocols and low-level complex vehicle dynamics [6]. While it is convenient to study each protocol separately, there is an eventual need for each CAV to be able to cooperate in multiple situations, with multiple cooperation protocols integrated into a social brain module.

4 Conclusion

CAVs can cooperate to improve safety and efficiency in a wide variety of situations. This paper has outlined the enormous possibilities for transforming transport with cooperative CAVs. However, there are also particular issues that need to be addressed via inter-disciplinary research for the benefits to be fully realized. The social brain will complement the close proximity sensing and intelligence that is already being built into CAVs today.


  • [1] UR:BAN Human Factors in Traffic: Approaches for Safe, Efficient and Stress-free Urban Traffic. Springer, 2018.
  • [2] A. Aliedani, S. W. Loke, A. Desai, and P. Desai. Investigating vehicle-to-vehicle communication for cooperative car parking: The copark approach. In 2016 IEEE International Smart Cities Conference (ISC2), pages 1–8, Sept 2016.
  • [3] Shunsuke Aoki and Ragunathan (Raj) Rajkumar. A merging protocol for self-driving vehicles. In Proceedings of the 8th International Conference on Cyber-Physical Systems, ICCPS ’17, pages 219–228, New York, NY, USA, 2017. ACM.
  • [4] J. Axelsson. Safety in vehicle platooning: A systematic literature review. IEEE Transactions on Intelligent Transportation Systems, 18(5):1033–1045, May 2017.
  • [5] Nesrine Bessghaier, Mahdi Zargayouna, and Flavien Balbo. An Agent-Based Community to Manage Urban Parking, pages 17–22. Springer Berlin Heidelberg, Berlin, Heidelberg, 2012.
  • [6] J. Campbell, C. E. Tuncali, P. Liu, T. P. Pavlic, U. Ozguner, and G. Fainekos. Modeling concurrency and reconfiguration in vehicular systems: A -calculus approach. In 2016 IEEE International Conference on Automation Science and Engineering (CASE), pages 523–530, Aug 2016.
  • [7] Jieqiong Chen and Guoqiang Mao. On the security of warning message dissemination in vehicular ad hoc networks. Journal of Communications and Information Networks, 2(2):46–58, Jun 2017.
  • [8] Prajakta Desai, Seng Wai Loke, Aniruddha Desai, and Jugdutt Singh. CARAVAN: congestion avoidance and route allocation using virtual agent negotiation. IEEE Trans. Intelligent Transportation Systems, 14(3):1197–1207, 2013.
  • [9] M. Jenkins, D. Duggan, and A. Negri. Towards a connected bicycle to communicate with vehicles and infrastructure : Multimodel alerting interface with networked short-range transmissions (main-st). In 2017 IEEE Conference on Cognitive and Computational Aspects of Situation Management (CogSIMA), pages 1–3, March 2017.
  • [10] Dongyao Jia, Kejie Lu, Jianping Wang, Xiang Zhang, and Xuemin Shen. A survey on platoon-based vehicular cyber-physical systems. IEEE Communications Surveys and Tutorials, 18(1):263–284, 2016.
  • [11] Maryam Kamali, Louise A. Dennis, Owen McAree, Michael Fisher, and Sandor M. Veres. Formal verification of autonomous vehicle platooning. Science of Computer Programming, 2017.
  • [12] Evangelia Kokolaki, Merkouris Karaliopoulos, and Ioannis Stavrakakis. Opportunistically assisted parking service discovery: Now it helps, now it does not. Pervasive Mob. Comput., 8(2):210–227, April 2012.
  • [13] Yannis Labrou, Tim Finin, and Yun Peng. Agent communication languages: The current landscape. IEEE Intelligent Systems, 14(2):45–52, March 1999.
  • [14] C. Liao, J. Chang, I. Lee, and K. K. Venkatasubramanian. A trust model for vehicular network-based incident reports. In 2013 IEEE 5th International Symposium on Wireless Vehicular Communications (WiVeC), pages 1–5, June 2013.
  • [15] Seng W. Loke. Crowd-Powered Mobile Computing and Smart Things. Springer Briefs in Computer Science. Springer, 2017.
  • [16] Rodolfo I. Meneguette, Geraldo P. R. Filho, Daniel L. Guidoni, Gustavo Pessin, Leandro A. Villas, and Jó Ueyama. Increasing intelligence in inter-vehicle communications to reduce traffic congestions: Experiments in urban and highway environments. PLOS ONE, 11(8):1–25, 08 2016.
  • [17] Adam Millard-Ball. Pedestrians, autonomous vehicles, and cities. Journal of Planning Education and Research, 0(0):0739456X16675674, 0.
  • [18] E. Moradi-Pari, H. N. Mahjoub, H. Kazemi, Y. P. Fallah, and A. Tahmasbi-Sarvestani. Utilizing model-based communication and control for cooperative automated vehicle applications. IEEE Transactions on Intelligent Vehicles, 2(1):38–51, March 2017.
  • [19] Pablo Noriega and Dave de Jonge. Electronic Institutions: The EI/EIDE Framework, pages 47–76. Springer International Publishing, Cham, 2016.
  • [20] E. Ohn-Bar and M. M. Trivedi. Looking at humans in the age of self-driving and highly automated vehicles. IEEE Transactions on Intelligent Vehicles, 1(1):90–104, March 2016.
  • [21] John Searle. Making the social world: The structure of human civilization. Oxford University Press, 2010.
  • [22] G. Tasseron, K. Martens, and R. van der Heijden. The potential impact of vehicle-to-vehicle communication on on-street parking under heterogeneous conditions. IEEE Intelligent Transportation Systems Magazine, 8(2):33–42, Summer 2016.
  • [23] Spyros G. Tzafestas. Roboethics: A Navigating Overview (Intelligent Systems, Control and Automation: Science and Engineering) . Springer, 2016.
  • [24] J. Wei, J. M. Dolan, and B. Litkouhi. Autonomous vehicle social behavior for highway entrance ramp management. In 2013 IEEE Intelligent Vehicles Symposium (IV), pages 201–207, June 2013.
  • [25] G. Xiong, H. Li, Y. Jin, J. Gong, and H. Chen. Collision avoidance system with cooperative adaptive cruise control in highway entrance ramp environment. In 2017 18th International Conference on Advanced Robotics (ICAR), pages 549–553, July 2017.