A Future-Forward Proposal for a System of Streets and Autonomous Vehicles

City Science team


The impending introduction of autonomous vehicles (AVs) has posed regulatory and ethical questions regarding how they should operate.  Much of the previous literature on this subject has explored these questions with an underlying model of streets based on the present.

This project takes a different approach by putting forward a future vision for streets where privately owned and operated vehicles are no longer dominant and shared transit is more pervasive.  In doing so, this work expands the current discussions around individual AVs to the system of streets they will occupy.  It views the topology of streets and the rules that govern them, coupled with the vehicles that move through the streets, as an autonomous system, or machine. This project proposes updates to this autonomous system in order to build a more equitable system for a future where AVs will be ubiquitous. We present a design with two parts in order to ensure that AVs operate in the public’s best interests: 

  1. An update to the laws that govern the use of roads, vehicle regulations and safety standards.  
  2. A requirement that AV decision making code be open sourced.


Autonomous vehicles (AVs) will be ubiquitous in our future city streets [1].  Proponents of AVs advocate for their accelerated adoption, noting potential benefits such as reducing pollution and improving road safety [2].   Advanced AV technologies promise to be more reliable than the average human driver and may eliminate up to 90% of car crashes [3].   To this end, the U.S. Department of Transportation and the National Highway Traffic Safety Administration (NHTSA) have published federal guidance for autonomous vehicles [4].   The guidance encourages states to begin testing them on local streets, and states such as Massachusetts have already begun doing so in pilot programs [5].   However, the guidance provided is general and underspecified.   It leaves open many questions that will need more specific answers before AVs can drive beyond the current phase of limited tests and into public use. Many of these questions are regulatory.   Others are ethical. All are of concern to both government officials, automobile manufacturers, and any member of the public who uses the streets.  This project proposes an approach to resolving open regulatory and ethical questions while also leveraging the potential of AV technology to reframe the priorities and morality of our streets.

The introduction of AV technology presents the opportunity to upgrade how the system of streets serves the public.   If AV driving algorithms are designed to prioritize higher occupancy vehicles, bikes, and pedestrians, we can expect the increased use of shared transit and a gradual shift to a system of streets that are safer and more equitable.


Ethical dilemmas regarding AVs are commonly discussed among moral philosophers and machine ethicists.  The well known “trolley problem” (*1) is a thought experiment used in the research of ethics and moral psychology that has modern variations involving AVs.  Ethicists theorize over these scenarios that AVs should be able to morally handle before their adoption can be acceptable.

For example, consider the scenario described by technology ethicist Patrick Lin, where an autonomous car carrying a single passenger is driving on a narrow road.  Suddenly, a bus full of people appears around the corner, and the robotic car calculates that given the velocities and close proximities of the two vehicles, a crash is imminent.  This is a dilemma with two main options: 1) The car can brake to slow before the crash, risking the lives of all people in both the car and the bus; 2) The car can swerve off the road, maximizing the risk to its own passenger while sparing all bus passengers of potential harm [6].

A similar ethical dilemma was recently posed to the public.  Researchers [7] conducted a study where participants were presented with a scenario where a car must choose between killing several pedestrians, or its own passenger.  When asked which option would be the most moral way to program an AV, participants largely favored AVs that sacrificed their passenger in order to save the lives of the pedestrians.  The public’s preference for AVs to minimize overall harm when it comes to their general behavior suggests that AVs should behave in a way that philosophers would consider utilitarian (*2).

Nevertheless, when study participants were asked to report their likelihood of purchasing such passenger-sacrificing AVs versus AVs that would instead protect their own passengers, the participants preferred the self-protective models for themselves.  This purchasing preference may seem like an incentive for AV manufacturers to prioritize passenger safety in AV decision making algorithms over all else. And the dichotomy of utilitarian and self-serving interests may seem like a conflict that will impede the public acceptance and adoption of AVs, delaying the larger safety benefits they can bring to all.


However, there need not be such a conflict.  There is a logical flaw in using many of the social dilemmas that theorists and policy makers have anticipated for AVs to regulate or ethically guide the future.  The problem is that these social dilemmas (such as the one posed above [7]) are limited to a perspective based on our current system of streets and vehicles.  Presupposing the priorities and incentives of our current system can be problematic when trying to build for the future.

In our present system, cars are viewed as privately owned and operated vehicles for consumer purchase. With this perspective, it is easy to assume that an AV should have allegiance to its owner [6] and prioritize protecting the safety of its own passenger at all costs, including the cost of bringing harm to numerous others on the road.  This assumption does not readily transfer to a future system of shared vehicles. For example, consider public buses. Since buses are part of a transit system designed to benefit the public, we may assume that an autonomous bus would prioritize minimizing harm to the public, whether the public is riding in the bus, or in another bus, or walking across the street.  If the previously presented dilemmas for AVs are modified to replace the cars with public buses, the decision calculus changes. In scenarios with only buses and pedestrians, a public preference for utilitarian approaches that prioritizes the lives of many over the life of one may be clearer. In dilemmas where only shared vehicles or pedestrians are involved, there is no opportunity to consider and prioritize oneself as “the passenger” who must choose between a self-protective vehicle or a vehicle that maximizes public wellbeing.  In such scenarios, any one person is just as likely to be riding one vehicle versus another, versus walking as a pedestrian.

With this reasoning in mind, consider a future where AVs operate as shared transit, like trains or buses.  Whichever vehicle someone is riding in at any given time, or whether they are walking on the streets, is arbitrary.  In this future, the ethical driving dilemmas simplify. The public can more easily take what philosophers call the original position (*3), and agree on a utilitarian approach for AVs to prioritize minimizing harm to the public.

The impending AV technologies offer the opportunity to design for such a future.  Yet the impact to our road system that AV technologies will bring need not be limited to vehicles.  The ways that AVs are adopted and regulated will affect walking, biking, and the other modes of transit brought by future innovations (*4).  AV technologies can be leveraged to benefit society by properly designing around how they should interface with these other modes of transit.  For example, cities benefit when more people choose to walk or bike instead of commuting by private cars, because these commuting modes are healthier and more environmentally sustainable.  Unfortunately, pedestrians and bikers may be discouraged by how vulnerable they are on the streets due to the dangers cars pose. In the United States, a pedestrian is injured by an automobile every 8 minutes [8], and pedestrians are 1.5 times more likely than vehicle occupants to be killed in a car crash on each trip [9].  The proper design and introduction of AVs can make pedestrians and bikers safer and thus more common. AV driving technologies will be better than human drivers, who are at times negligent or hurried, and fail to follow best safety practices and the rules of the road.

The introduction of AVs also offers the opportunity to adapt streets to new priorities.  Driving technologies can be designed around priorities for shared transit and the safety of pedestrians and bikers.  Such a design paradigm would maximize the potential for AVs to benefit society once deployed at scale. Public safety and sustainability would be impacted, as well as commuting choices.  Streets that prioritized higher occupancy vehicles, making them more efficient and safer than private vehicles, would incentivize the use of shared transit so that the described future world, where most AVs are shared, is realized.  The streets of this future world would be more moral, not only because moral philosophers and the public could align on ethical driving dilemmas, but also because they would be driven by the public’s best interests.

A practical question is how to get to this future of AVs and reformed priorities, and how to rely on AVs operating in the public’s best interests.  This project proposes an approach towards attaining and facilitating such a future system with both regulations and open source code. The proposal views the topology of streets and the rules that govern them, coupled with the vehicles that move through them, as a system, or machine.  The proposed approach aims to advance the system of streets and vehicles towards a more “moral machine” than the present one.


Our network of roads and the rules that govern them enables the public’s freedom of mobility.  This network is used by people and vehicles every day. Its infrastructure, together with the patterns and movements along it, can be viewed from above as an autonomous system.  It is a system containing individual mobility choices that also form larger patterns when observed at scale.

The topological and legal infrastructure of our roads are the foundation for computer routing algorithms and human decisions about how to move from here to there.  The route for a car may be determined by physical factors, such as which roads are shortest and how the roads are connected, as well as governed factors, such as which roads are one-way or have stop signs or multiple lanes.  

The shortest and most efficient route is often chosen for cars.  Safety is a larger consideration for those on bicycles or foot, and their safe choices are limited.  Few US roads have dedicated lanes for bicycles, but bicyclists may be better off limiting their routes to them.  Likewise, pedestrian routes may be limited to where sidewalks and crosswalks are present. As such, the network of streets and its rules in which the public moves is car-centric, and those outside private vehicles often must take a backseat.

Our current system of streets embeds a prioritization for cars in both structure and governance that is inequitable.  Structural factors include how bike lanes and crosswalks are limited features of streets; the streets are otherwise the domain of vehicles.  There are also governed factors, such as right-of-way laws which determine when a vehicle is required to yield. Bicycles are considered “vehicles” by law, so they are subject to the same rules as car drivers; although bicycles are more vulnerable than cars, they are not generally granted right of way.  US federal and state right-of-way laws require cars to yield to other cars that arrive first to intersections. They only require yielding to pedestrians who are already in crosswalks, but not necessarily to pedestrians waiting to cross (*5). These requirements encode an underlying preference for cars. Crosswalks are intersections for pedestrians, yet pedestrians must wait at intersections while other vehicles are given preference.  Even when bicyclists and pedestrians do have legal right of way, they still have a practical need to yield to vehicles; vehicles are more powerful. In the case of a collision, a car’s driver will likely be safe while a bicyclist or pedestrian will suffer bodily harm. Whether they had legal right of way will matter little compared to the harm sustained by them.

The US public relies on the system of streets for mobility, but the current system that prioritizes privately operated cars is not serving the public equitably.  A system that prioritizes and incentivizes shared transit, biking, and walking can be a more moral one with social impact along 3 main avenues:

  1. Safety: A system that prioritizes the safety of higher occupancy vehicles and the most vulnerable on the streets (pedestrians and bikers) will see fewer fatalities on the roads.
  2. Sustainability: When people transition from personal vehicles to shared transit, bikes, or walking, there is less pollution from cars.
  3. Equity: More shared transit enables those who otherwise have limited access to personal automobiles or mobility options.

Although there are public shared vehicles, namely buses, they are often under-resourced.  Increasingly, private car sharing services, such as Uber and Lyft, have filled a void. The continued success and expansion of these companies is a testament to the public’s growing appetite for shared vehicles.

The increasing use of shared vehicles, whether supplied by public transit agencies, or private car sharing services, can be furthered by the design of mechanisms that prioritize higher occupancy vehicles on the road.  These mechanisms can be implemented within AV driving software along with other rules that benefit bicyclists and pedestrians. The result will be that as new AVs drive into our future streets, the new priorities embedded in their driving software will as well.

As the public begins thinking about vehicles as shared goods, their expectations regarding how vehicles should be regulated and the rules that guide them will change.  A public that views vehicles as shared goods will be more likely to favor utilitarian rules and regulations for the vehicles. At the same time, rules and regulations for AVs can act as mechanisms to shift how vehicles are used and help facilitate a public change in perception of vehicles as private commodities to shared goods.  For example, rules that cause a vehicle to yield to higher occupancy vehicles will make shared transit more efficient, safer, and thus more appealing. In this way, changes in public perception of vehicles, and changes to rules to match the public’s changing expectations for vehicles, can be interrelated and cyclical.

Described below is a two-part design to guide such a cycle towards the more moral system of streets envisioned for the future:

  1. An update to laws for the streets and vehicle regulations in order to prioritize high occupancy vehicles, bicyclists, and pedestrians.  These rules will guide the design of algorithms that drive AVs.
  2. A requirement that all AV driving algorithms be open source.  This will enforce a transparent implementation of the rules and regulations.


Traffic rules and vehicle regulations can be a bridge between the public’s expectations and AV manufacturers that must build and program vehicles to legally drive by these rules and regulations.  They should reflect public decisions regarding how AVs should drive, and be clear enough so that an understanding of the rules programmed into AV driving software is accessible through their policy.

Rules and regulations present a unique opportunity to change streets when applied as programmed rules for AVs.  New human drivers may have difficulty memorizing the safety laws and rules in their driver’s manuals. Yet computer systems, such as those found in AVs, are ideal for storing rules in their memory and following them with exactness.  A law to facilitate road safety and fairness such as “stop at a stop sign” may not always be followed by negligent or hurried drivers, but an AV that runs through a stop sign has a bug.

A design principle for rules and regulations for AVs should the be the prioritization of higher occupancy vehicles, walking, and biking.  In the same way an AV should be expected to see and stop for a stop sign on the side of the road, it should be expected to see and stop for a pedestrian waiting at a crosswalk.  Laws such as the previously provided right-of-way law for intersections can also be reconsidered. This law can be interpreted as an algorithm with a set of clauses that determine when a car should yield to others: when another car has already arrived to the intersection, when another car is to the right, etc.  Right-of-way laws can be updated to insert clauses that prioritize shared transit, pedestrians, and cyclists by requiring privately operated AVs to yield to them.

Rules and regulations can also be designed to address the driving scenarios that are not  as simple as yielding at intersections. The ethical driving dilemmas that ask whose life should be put in danger and whose should be spared may seem like overly constructed or rare scenarios, but they can be useful thought experiments [6] to tease out moral principles and priorities to apply to other, less extreme, scenarios.  For example moral principles that guide how AVs should operate around high-occupancy vehicles, pedestrians, and bikers, might govern how closely an AV should drive alongside a biker or pass a pedestrian, or how an AV should maneuver around another vehicle with more occupants or yield to the other vehicle. These are not extreme examples of life or death, but they still include elements of risk that AVs will need to negotiate on the road in order to respect the lives of others and minimize the potential for harm.

There need not always be “hard coded” rules to navigate AV driving decisions.  Many computer algorithms and computational models are guided by optimization problems (*6).  The driving algorithms of future AVs can involve optimization problems that are designed around moral principles and the priorities regulated by law.  For example, if rules or regulations state that AVs should prioritize the safety of multiple people over the safety of a single passenger, or the safety of a biker over an AV’s own efficiency, then AV driving choices should be optimized to do so.

Legislatures determine the laws that govern the road, but there are also US federal safety standards and regulations for automobiles.  The Federal Motor Vehicle Safety Standards (FMVSS) specify minimum design, construction, performance, and durability requirements for motor vehicles and items of motor vehicle equipment.  They are developed and enforced by the National Highway Traffic Safety Administration (NHTSA) so "that the public is protected against unreasonable risk of crashes occurring as a result of the design, construction, or performance of motor vehicles and is also protected against unreasonable risk of death or injury in the event crashes do occur." [10].  NHTSA tests how well vehicles meet the FMVSS with performance based tests. For example, crash tests for FMVSS No. 208 measure how well a passenger vehicle would protect its occupants in the event of a serious frontal crash. The “Full Frontal Fixed Barrier Crash” test (or “Rigid Barrier” test) involves driving a vehicle into a head-on rigid obstacle at varying velocities [11].

These performance based tests were deemed important given that crashes often occur.  Other Federal Motor Vehicle Safety Standards are design requirements for crash avoidance (e.g. FMVSS No. 111 requires rearview mirrors).  RAND corporation, an American nonprofit policy think tank, has pointed out that AV sensors and driving intelligence will necessitate updates to the current standards.  RAND recommends continuing to use performance based tests to assess AV safety, and also proposes tests to assess whether AVs can safely avoid crash scenarios [12].  It proposes both on-road tests and computer-based simulation tests.  Tests can assess how well an AV avoids collisions with obstacles, including other vehicles, pedestrians, and bicycles.

An AV’s ability to adhere to new rules and regulated driving priorities can also be assessed in performance based tests, and these tests can be conducted both on the road and in simulations.  Simple tests might include the following:

  • Whether the AV stops for a pedestrian waiting at a crosswalk.
  • When passing a bicycle on one side and a car on the other side, whether the AV gives the bicycle more space by X amount, or otherwise does not pass.
  • Whether the AV yields at an intersection to shared transit vehicles.

These are only simple examples to show how performance based tests can assess how AVs navigate regulated driving priorities.  Assuming AVs will have mechanisms to detect whether other vehicles are low or high occupancy vehicles, these tests can assess how readily a low occupancy AV vehicle yields to higher occupancy vehicles.

The advantage of using performance based tests is that regulating the design of AV driving algorithms and code can then be avoided.  Regulating around performance instead of design is preferable because technology tends to evolve more quickly than public policy.  New innovations in AV technologies should not be hindered by outdated policy regulating technical designs if new innovations lead to outcomes that are as good, or better, than the previous versions.

Performance tests likely will not be sufficient to fully establish public trust in AVs.  In order to establish accountability for AV manufacturers, an update to rules and regulations should be accompanied by a requirement that all AV driving decision code be open source.


Open source code invites public audit from both government organizations and interested individuals.  Examination of the code can best determine whether AV driving algorithms are designed to optimize for established priorities and public expectations.  There are many ways that moral principles and regulated priorities can be implemented in AV software. Making this software open source will allow AV developers to choose their own designs and remain accountable to the public for how well it meets public standards and expectations.

Sarah Thornton and her team at Stanford’s Dynamic Design Lab have already demonstrated such a design for elements of AV driving software.  Thornton, et al., used a “value sensitive design” approach to develop an algorithm to control an AV’s speed, where the algorithm optimized for human values [13].  These values were chosen with respect to the safety and convenience of both the AV’s passengers and pedestrians. The values were represented in cost and reward functions for their AV’s speed control algorithm to optimize for, and the final implementation has been tested on a road course with crosswalks and pedestrians.  Value sensitive design is an iterative methodology that formalizes connections between human values, or priorities, to engineering specifications. The methodology was used in the construction of a partially observable markov decision making process for this particular implementation of an AV speed control algorithm.

There are many other approaches to optimize AV decision making software for values, priorities, and rules that match public expectations.  AV manufacturers should not be concerned that making their implementations open source will reduce their competitive advantage; the open source code is only a small component in the larger product they are marketing and distributing.

A common concept in computer programming is “modularity”.  Modularity is about separating components. For example, the software that causes vehicle tires to spin based on the vehicle’s desired speed is separate from the software that determines the desired speed.  The code for AV driving decisions can be treated as a component, or module, that interfaces with other parts of a vehicle’s system. These other parts may be system components that detect and interpret the vehicle’s surroundings, or the components that cause the vehicle to turn, etc.

Only the code that determines AV driving decisions need be open source.  The software for other components can remain proprietary so that vehicle manufacturers can continue to privately optimize the parts of their product that do not have ethical implications.  While AV manufacturers implement and publicize software designs that meet the public’s priorities and standards, they can continue to compete for a competitive advantage.


Open source code provides transparency and a means to ensure that AVs operate in the public’s best interests.  Coupling a requirement for open source AV driving algorithms with the proposed adjustments in rules and regulations can shift public mobility patterns.  Private vehicles will seem less advantageous when a fleet of AVs are programmed to prioritize higher occupancy vehicles, bicycles, and pedestrians. As the roads become an environment where shared transit is more advantageous, incentives will shift from the use of private to shared vehicles.  An increased market demand for shared transit may be fulfilled by public agencies or private industry, such as the car sharing services that have already entered the market. With moral principles embedded in the algorithms of vehicles, and with priorities on the road shifted to incentivize shared transit, the system of streets can transition to a more “moral” system.

The two-part design of (1) updates to rules and regulation, and (2) a requirement for open source driving algorithms, was presented as an approach to facilitate a transition to a more “moral” system of streets.  There can be a future system of streets where more vehicles operate as shared transit, and where the rules of the streets, and the algorithms that drive vehicles, align with public interests. AVs can be tools to drive forward a better system of streets, if they are properly introduced.


A simulation, with two alternative futures for AVs, supplements this written work (with all code open source).  In one version of the simulation, AVs operate under the present model of private vehicles. In the other version, AVs are shared vehicles that follow the rules outlined in the regulatory portion of the proposed design.  The two versions show a contrast in mobility patterns, congestion, and use of transit by bicycles and walking.

For video of the simulation, open source code, and more details, see:


This technology project is a fork of work done by the City Science team at MIT’s Media Lab. It originated as an interactive physical table + simulation for an exhibition at the Cooper Hewitt Museum.

Much credit goes to: (in alphabetical order) Ariel Noyman, Arnaud Grignard, Carson Smuts, Gabriela Advincula, Guadalupe Fernandez, Kent Larson, Luis Alberto Alonso Pastor, Maitane Iruretagoyena, Margaret Church, Markus ElKatsha, Ronan Doorley, Yan Zhang, Yasushi Sakai



(*1) The general form of the “trolley problem” is commonly presented as follows:

There is a runaway trolley on a track that branches.  On the main branch of the tracks, there are five people tied to the tracks.  On the other branch there is one person tied to the tracks. The trolley is currently on the main branch and will hit the five people unless you pull a lever to switch the tracks and redirect the trolley to hit the one person on the other branch.  You have two options:

  1. Do nothing and allow the trolley to kill the five people on the main track.
  2. Pull the lever, diverting the trolley onto the side track where it will kill one person.

Which is the more ethical option?

(*2) Utilitarianism is an ethical theory that considers the morally right action the one that maximizes utility, or brings about the great amount of good for the greatest amount of people.

(*3) The original position is a fair and impartial view.  It was developed by philosopher John Rawls as a thought experiment where those who must make decisions about the structure and function of a society do not know what roles they will play in that society, or what characteristics they will have, such as race, sex, status, or income.  Without this information, participants are expected to make decisions impartially and rationally.

(*4) For example, researchers at the MIT Media Lab are developing the “Persuasive Electric Vehicle” (PEV) as a shared-use, autonomous, lightweight, electric tricycle.  It is designed to alternate between carrying passengers and cargo based on the public’s needs, and it can be summoned with an app. See

(*5) For example, see Massachusetts legislature Part I, Title XIV , Chapter 89, Section 11: Marked crosswalks; yielding right of way to pedestrians; penalty.  “When traffic control signals are not in place or not in operation the driver of a vehicle shall yield the right of way, slowing down or stopping if need be so to yield, to a pedestrian crossing the roadway within a crosswalk marked in accordance with standards established by the department of highways if the pedestrian is on that half of the traveled part of the way on which the vehicle is traveling or if the pedestrian approaches from the opposite half of the traveled part of the way to within 10 feet of that half of the traveled part of the way on which said vehicle is traveling.”

(*6)  In mathematics and computer science, optimization problems are used to find the “best” possible solution to a problem with multiple solutions.  Optimization problems have objectives, variables, and constraints. In an overly simplified example for AVs, an objective might be to minimize the time it takes for a vehicle to reach its destination.  Variables might include speed, and constraints might include speed limit.


[1] National Highway Traffic Safety Administration (NHTSA). 2018, Automated Vehicles for Safety. Retrieved from NHTSA website:

[2] Spieser, K., et al., Toward a Systematic Approach to the Design and Evaluation of Automated Mobility-on-Demand Systems: A Case Study in Singapore. In Road Vehicle Automation, 2014. pp. 229-245.

[3] Smith, B. W. 2013, Human Error as a Cause of Vehicle Crashes, The Center for Internet and Society at Stanford Law School.

[4] U.S. Department of Transportation. 2018.  Automated Vehicles 3.0: Preparing for the Future of Transportation.

[5] City of Boston 2018. Autonomous Vehicles: Boston’s Approach. Retrieved from City of Boston website:

[6] Lin, P. 2016. Why Ethics Matters for Autonomous Cars. In: Maurer M., Gerdes J., Lenz B., Winner H. (eds) Autonomous Driving. Springer, Berlin, Heidelberg.

[7] Bonnefon, J.; Shariff, A.; and Rahwan, I. 2017. The Social Dilemma of Autonomous Vehicles. Science, 352(6293): 1573-1576.

[8] Rosen E, Sander U. Pedestrian fatality risk as a function of car impact speed. Accid Anal Prev 2009;41:536-542.

[9] Beck LF, Dellinger AM, O’Neil ME. Motor vehicle crash injury rates by mode of travel, United States: Using exposure-based methods to quantify differences. Am J Epidemiol 2007;1

[10] National Highway Traffic Safety Administration (NHTSA). 1999. Federal Motor Vehicle Safety Standards and Regulations.  As of October, 2018:

[11] Office of Vehicle Safety Research. 1999. Updated Review of Potential Test Procedures for FMVSS No. 208.  Prepared for the National Highway Traffic Safety Administration (NHTSA).

[12] Fraade-Blanar, L., and Nidhi, K. 2017.  Autonomous Vehicles and Federal Safety Standards: An Exemption to the Rule? RAND Corporation.

[13] Thornton, S. M., Lewis, F., Zhang, V., Kochenderfer, M., and Gerdes, J. C., Value Sensitive Design for Autonomous Vehicle Motion Planning. In IEEE Intelligent Vehicles Symposium, 2018.

[14] The General Court of the Commonwealth of Massachusetts, General Laws, Part I, Title XIV.  2018.

[15] Freeman, Samuel, "Original Position", The Stanford Encyclopedia of Philosophy (Winter 2016 Edition), Edward N. Zalta (ed.),

Related Content