3.1 A new urban mobility in the 15 minutes walkable city
3.1.1 A new framework for urban mobility - From nature towards a shared and collaborative mobility system of multi-functional, electric and autonomous light-weight vehicles
The concept of future mobility as multi-functional vehicles behaving as a bio-inspired collabor- ative system combines, expands, and translates ideas from nature, swarm robotics, and vehicle platooning. Natural swarm behaviors have served as an inspiration for researchers in robotics, leading to the field of swarm robotics [16]. While the application is different from mobility, in the field of swarm robotics there are excellent examples of translating behaviors from natural to artificial systems.
Figure 3.1: Example of mutualism in nature between the Nile Crocodile and the Egyptian plover bird. [26]
Within the field of mobility, research in vehicle platooning studies how vehicles can travel in coordination [14]. While the CS Group’s view aims to extend interactions between vehicles from coordination to various ways of collaboration, vehicle platooning can be seen as a first step towards this future.
This section explains how some of the concepts of swarm robotics might be transferred into the field of mobility and combined with vehicle platooning. The characteristics and benefits of these models are also outlined in this section, and, in doing so, provide details are provided on the characteristics and potential benefits of the proposed future mobility framework.
3.1.1.1 From swarm robotics to urban mobility
The term ’swarm’ in the context of robotics was used for the first time by Fukuda et al. [95] and G. Beni [96] in 1988. Fukuda studied how to design a robotic system that, having an intelligence analogous to the biological gene, would be able to dynamically reorganize its shape and structure for a given task by employing limited available resources [95]. G. Beni, instead, defined what he named as Cellular Robotic System (CRS).
The systems considered consist of a large (but finite) number of relatively simple robotic units capable of accomplishing, collectively, relatively complex tasks. ...
The system formed by the autonomous robotic units called Cellular Robotic System or CRS is characterized by its reliability and its ability to self-organize and self- repair. [96, p. 57].
In 1993, G. Beni and J. Wang [97] argued the unpredictability in the behavior of systems that are capable of producing order can result in a non-trivial, different form of intelligence, which was named as ’swarm intelligence’. Swarm intelligence is a form of artificial intelligence inspired by insect colonies, emerging from the collective association of individual agents behaving in a way that goes beyond the aggregation of the individual capabilities of each agent [87].
The field of swarm robotics implements swarm intelligence into multi-robot systems [87]. Trying to emulate the natural behavior of social insects, researchers in this field have pursued imitating behaviors like foraging, flocking, sorting, stigmergy, or cooperation [98], which are based only on local information [99]. Based on the same principle, swarm robotics aim to construct robust, flexible, and scalable systems by using simple robots and local interactions [100].
Swarm robots are characterized by some properties such as simplicity, self-organization, de- centralization, task division and local interactions. These characteristics lead to scalability, robustness and flexibility, benefits that are also critical to urban mobility systems. Both these properties and benefits have been defined by G. Beni [101] and completed by others [16,102,103].
The following sections provide an overview of these characteristics (which could be understood as benefits in some cases) indicating how they could impact mobility.
Self-Organization
In the same way that biological swarms self-organize into patterns, robotic swarms show intelli- gence by producing ordered structures in an unpredictable way [101,104]. Self-organized systems in mobility can outperform traditional approaches based on fixed routes and schedules [105] and self-organization could solve some of the challenges inherent to vehicle networks [106].
Decentralization
Robots act by themselves when performing tasks with a distributed control topology and, con- sequently, any swarm member can make decisions independently [102]. In connected mobility, decentralization can be critical to increase security and robustness, decreasing delays in com- munication, and solving the problem of a single node failure, as has been proved in research about network topologies [107]. Decentralization could also improve scalability; having self- coordinating agents without a centralized communication allows having a greater number of agents [87].
Local Interactions
Inter-robot communications and sensing are limited to be only local; a property directly inher- ited from natural swarms, which heavily rely on local interactions for cooperation [102, 103].
Local interactions lead to benefits such as flexibility, and robustness [104]. In addition to these advantages, a system based on local interactions can be very scalable by dynamically adapting to different fleet sizes without any change in the software or hardware, which is very relevant for real-world applications such as urban mobility [102].
Task division
Robotic swarms act as a massive parallel computing system that, being able to create various solutions through task division between robots [108], can carry out more complex tasks than the
individual itself [102]. This improves efficiency and performance in task completion as compared to single robotic systems [109].
Simplicity
Agents in a swarm are simple, which has been defined as having a limited capability relative to the global task [87]. Additionally, the robots are usually quasi-identical, which allows for standardization and cost efficiency [16]. Current autonomous vehicles are much less simple and more capable relative to the task than swarm robots; however, working in swarm systems could potentially decrease the level of complexity of autonomous vehicles.
Scalability
Thanks to local interactions and self-organization, robotic swarms could theoretically count on an unlimited number of members [110]. In practice, robotic swarms could be composed of thousands, or even millions of units [101]. While clusters of vehicles might be formed by a smaller number of vehicles, collaborative systems of two or a few agents could already show many of the characteristics seen in larger swarms, such as task division, decentralization, or local interactions, and so show some of its benefits [111].
Robustness
The number of agents in a swarm must be large enough to provide redundancy and robustness so that swarms can continue working even after the failure of some individuals [101]. Robustness is also fundamental in transportation. Numerous studies have shown the relevance of reliability in mobility mode choice behavior [106, 112], showing that unreliability results in a negative effect in mode choice for commute trips, even for those who have flexible work schedules [113].
Moreover, in the event of disasters, robustness in transportation is critical since it provides access to emergency supplies and services [114], and it is also needed in order to restore other services such as water, electricity, or communications [115].
Flexibility
Task division provides either optimization eliminating redundant efforts or extra security through redundancy [16]. For instance, a cluster of autonomous vehicles traveling together can share the navigation tasks required for autonomous driving, provide redundancy in the tasks that are more critical, or enable the vehicles to increase their performance thanks to the combined computational and physical capabilities.
In light of the above, there seems to be great potential in extending these concepts from swarm systems to mobility. The synergies of a future collaborative mobility would bring human behavior closer to natural systems, which have evolved through millions of years of evolution and are clear evidence of the power of collaboration [116].
3.1.1.2 From vehicle platooning to collaboration
Autonomous vehicles can communicate with other vehicles, humans, and the infrastructure and have the intelligence to make decisions based on this communication [71]. These communica- tion skills open the possibility for autonomous vehicles to cooperate, giving birth to concepts as vehicle platooning [117]. Maiti et al. define vehicle platooning as a closely following mechan- ism that allows vehicles to travel in a coordinated way, without any mechanical linkage, while maintaining a safety distance [14].
The concept of vehicle platooning has its origins in the 1960’s [118]. It started to gain attention when, in 1995, a research report related to the California PATH project showed promising results related to drag force reduction [119]. One of the most striking early demonstrations of these concepts took place years later, in the 2011 Grand Cooperative Driving Challenge (GCDC) [120], showing vehicles dynamically cooperating both in urban and highway scenarios.
Vehicle platooning has three fundamental elements: 1) Vehicle-to-Vehicle (V2V) communication that allows vehicles to coordinate with minimal resources [121, 122], 2) distributed control as a robust framework that allows each vehicle to decide how to act as a combination of the information received by the on-board sensors and V2V communication [122,123], and 3) Vehicle- to-Infrastructure (V2I) communication which allows vehicles to interact with the infrastructure in a variety of scenarios such as finding a charging spot, or navigating to attach to a platoon [124].
Platooning has multiple potential benefits, some of which are summarized in this section. Even if platooning research has mainly focused on heavy vehicles, extending this concept for lighter vehicles could lead to similar benefits [121]. Since platooning (i.e., coordinated travel) can be seen as a first step towards collaboration, the framework presented would inherit the benefits of platooning. These benefits are analyzed in the following sections.
Fuel Consumption
The main research line in vehicle platooning is related to heavy vehicles due to direct benefits in fuel consumption associated with drag reductions [123, 125]. Drag reductions lead to environ- mental and financial gains, which is very relevant as fuel is the highest cost for a heavy vehicle fleet owner [122].
Traffic flow efficiency
Systems based on velocity and space regulation such as the Intelligent Cruise Control System (ICC) have been proposed to increase traffic flow efficiency [126]. For instance, in an early study Michael et al. [127] quantified the increase in traffic flow efficiency, concluding that platoons formed by up to ten light-duty passenger vehicles could double or even triple highway capacity.
However, must be noted that several studies indicate that, even if connected autonomous vehicles might increase traffic flow efficiency, this would, in turn, induce travel demand [128–130].
Safety
Among all traffic accidents, human error is involved in 50% to 90% of them [131]. It is predicted that autonomous vehicle platooning will contribute to road safety with systems such as Cooper- ative Adaptive Cruise Control (CACC); a system in which vehicles constantly communicate and measure distances to their predecessors [132, 133]. Similarly to how driver assistance functions are becoming each time more advanced, CACC systems are also evolving towards more advanced predictive safety systems [134, 135].
The future mobility that is presented in this proposal extends the interactions between vehicles from the coordination proposed by platooning models to a broader set of interactions. Being an extended version of platooning, it would inherit the aforementioned benefits could potentially reduce fuel consumption, increase traffic flow efficiency and improve safety.
In addition to expanding the possible interactions between vehicles, this model also proposes a different approach to the way platoons would be formed. Most platooning research considers centralized planning approaches but it has been argued that platoons might be formed ad-hoc between drivers who do not necessarily know each other [136]. Since in this proposed framework
clusters of vehicles would be formed by vehicles that might even be from different fleet owners, it would also require decentralized planning strategies that would enable collaboration between vehicles to happen spontaneously and without previous planning, based on local interactions.
Lastly, while most platooning research considers that all the agents are identical or at least very similar [137], proposal introduces the possibility of collaborating across scales and vehicle types forming heterogeneous clusters. Vehicles could have specific duties such as charging or providing maintenance services to other vehicles. Taking this idea even a step further, vehicles could also different levels of autonomy as well; for instance, vehicles with outdoor driving capabilities could guide vehicles that can only navigate indoors from one building to another.
All in all, this proposal for future mobility builds on top of the research in vehicle platooning by bringing concepts from swarm robotics that expand interactions between vehicles from com- munication to a broader set of interactions that imitate the behavior found in natural swarms.
Consequently, the potential benefits would also include those of swarm robotics and vehicle platooning.
3.1.2 Lane following algorithms to build driverless skills a lightweight mobility prototype
With the objective of developing those aforementioned ultra-lightweight vehicles, able to collab- orate in this future mobility, progress was made with the MIT Autonomous Bicycle. In order to showcase how this bike could circulate autonomously in cities, a first iteration of lane following skills were developed. Similar projects were researched to build a first iteration of a lane follow- ing algorithm. Two projects stood out as suitable candidates for inspiration for their simplicity and proved performance:
• The first analyzed project [27] was developed as a project for the Udacity Infosys Connect Self driving Car Nanodegree. It performs lane detection as well as object detection (e.g.
cars, bicycles). The first part of the project is focused on image distortion, which is key when working with computer vision.
Figure 3.2: Lane detection project for the Udacity Infosys Connect Self driving Car Nanodegree.
[27]
• The second analyzed project [28] constitutes a tutorial to learn how to develop a reduced size driverless car, able to follow a lane and recognize objects such as toys, traffic signals,
etc. It explains the whole pipeline from the acquisition of the hardware until having a complete operative system, step by step, with a clear outline of the algorithm.
Figure 3.3: Lane detection project to construct a reduced size driverless car [28]
Both projects constituted a solid base to build the first iteration of a lane following algorithm.
The second project, more complete and with a more understandable code, was chosen as the starting point.
3.1.3 The need of a system for dynamic V2V power-sharing in collaborative fleets of autonomous vehicles
A future in which shared, multi-functional, autonomous vehicles will behave as a bio-inspired, collaborative, system [18] opens the door to new approaches to vehicle charging. Based on this framework, vehicles could share battery with other vehicles while moving, doing what is more efficient for the system as a whole. This section provides an overview of current EV charging strategies and energy-transmission coupling solutions proposed in other transportation fields.
From this analysis, it is concluded that the ideal charging strategy for this proposal has the following characteristics: A) It allows for dynamic power transfer between vehicles, B) The energy transmission is wireless, C) Vehicles are physically coupled through an electromagnet.
3.1.3.1 EV charging strategies
Currently, most of the vehicles are charged with plug-in connections to the grid [138]. Plug-in charging is a mature technology, but a full charge can take 4-8h with a regular charger, 1-3h with a fast charger, and even with the most advanced chargers, it takes 30 min to charge the battery to an 80% [72]. In addition to being a time-consuming process, chargers are usually difficult to access, and users can have trouble finding free slots [72, 139]. As a result, charging stops represent a large portion of the trip time [140].
Overall, the sparse infrastructure and range anxiety have a detrimental impact on the adoption of EVs [139,141]. In addition to that, the cost of plug-in charger infrastructure is very high and, for example, in the United States, the installation costs are three to five times higher than the cost of the infrastructure hardware [142]. Due to the cost and inconvenience of plug-in charging solutions [139, 142], researchers and companies have analyzed numerous alternative models.
Figure 3.4: Mobile charger solution “Mobi EV Charger” by Freewire. [29]
For instance, to reduce the size of charging stations, some models propose vehicle-to-vehicle (V2V) charging, which allows multiple cars to be connected to a single connection [139,143,144].
While initially V2V charging was proposed for stations, Huang et al. proposed an approach based on mobile chargers [145], leading to commercial solutions such as FreeWire’s Mobi EV Charger [29]. This proposal reduces the need for infrastructure and eliminates the problem of the availability of connectors at each station, which can be especially critical in remote areas [146].
However, these proposals do not reduce charging times.
Conversely, other proposals are focused on reducing the charging times. For instance, in battery swap stations, uncharged batteries are almost instantly replaced by charged ones [76]. The swapping process can take around 5 minutes, which is the same time that it currently takes to refuel an ICEV [147]. However, the infrastructure for battery swap stations has a higher cost than plug-in charging [147] and the real-life application is limited because it requires batteries to be standardized [72, 80].
With the combined goal of reducing stop times to zero and removing the need to travel to a specific site, dynamic charging solutions propose to charge vehicles utilizing charging lanes [79, 82, 148–150]. Most proposals include wireless technology [83], but some have also proposed the use of a pantograph system [77] which has been implemented by Siemens [81]. Although the benefits of these dynamic charging technologies are promising due to increased driving range, decreased battery size, and having no stop time, they are infrastructure dependent, and the required infrastructure has a high cost and complexity [81, 151].
Finally, some researchers have proposed a model for dynamic V2V charging at a theoretical level. For example, in 2018, Zhang et al. propose to combine charging lanes with in-route wireless V2V energy transmission [151]. More recently, Nezamuddin and Dos Santos have also proposed a similar concept [140]. However, to the best of the CS Group knowledge, there is no practical implementation of this system [152]. Therefore, this proposal develops these ideas further, integrating them in a scenario in which autonomous vehicles will be part of collaborative fleets. Moreover, it proposes a technical solution that could enable V2V energy transfer on the go, which has been validated with a proof of concept prototype.
Dynamic V2V energy transfer is proposed as a method to charge a collaborative fleet of autonom-
Figure 3.5: Scheme of a catenary power system and the drivetrain of an overhead catenary truck. [30]
ous vehicles. In this project’s proposal, charging infrastructure would only be localized in a few strategic points. Vehicles would get charged in these locations and collaboratively distribute it to the rest of the vehicles in the system in the most efficient way. It is a solution that requires lightweight infrastructure, and that provides great flexibility, eliminates the problem of access- ing stations, and reduces cost. At the same time, it optimizes the usage of the vehicles at a system level, with less time and kilometers traveled related to charging. Finally, since vehicles can charge while moving, vehicle downtime is reduced, consequently increasing vehicle usage.
Finally, both of the previously mentioned articles [140, 151] present V2V in-route power-sharing at a very theoretical level and do not detail how this could be realized in practice.
3.1.3.2 Technologies for EV charging
As for today, most commercial vehicles use conductive charging methods to charge their batteries with plug-in solutions [138] [152]. Conductive plug-in charging is a mature and high-efficient technology [139]. However, it presents some disadvantages: The degradation of the cables over time can end in current leakage [153], and the connectors are an issue too, since different solutions lead to the need for adapters, increasing the cost of EVs and introducing safety risks [72]. In addition, the infrastructure is fixed in specific locations, it requires vehicles to be stationary while charging, and it has to be manually engaged, which is inconvenient, especially in adverse weather [139].
Extrapolating the railway solution for the energy supply to EVs, some researchers proposed the use of pantographs. This model is being proposed mainly for long-haul transportation with heavy-duty trucks in a combined approach; in routes with catenaries, vehicles can get energy supply while driving, while in paths without catenaries, they use their own battery [30]. The main drawbacks for this system are that there is still a need for infrastructure in some paths and that the cost of the infrastructure is high [81].
Wireless technology is another primary charging method [139]. Wireless transfer refers to trans- mitting electrical power through the air [152]. The leading wireless charge modes for EVs are capacitive power transfer (CPT) and inductive power transfer (IPT) [72]. IPT is the current most used wireless technology because it works with various gap distances [72]. It is being de- veloped by multiple researchers all over the world [153, 154]. IPT systems present high reactive power, which requires a compensation circuit and results in lower efficiency [138]. IPT also re- quires a ferromagnetic material, such as ferrite, to guide the flux, and this flux is very sensitive