Issues in Managing IT Infrastructure for Autonomous Transportation

Issues in Managing IT Infrastructure for Autonomous Transportation

Final Term Paper

July 31, 2019

Prof. Jerrard Gaertner 

MT8945 – MBA-MTI-TRSM, RU 

Abstract

 

This paper talks about the issues in managing IT infrastructure for autonomous vehicles (AV) specifically focussing on the challenges associated with the management of sensor technology and cybersecurity. The issues are looked at from the perspective of a CIO of a company in the AV industry.  In order to adhere to the word limit and to keep the research focussed, the paper focuses on Cars as a representative of the AV industry; while, cars can be a good proxy for commercial vehicles, they may not cover all issues applicable to other kinds of AVs such as automated trains. The paper finds that today’s car companies are addressing the need for a revamp in the IT infrastructure that is required to address the transition to AV manufacturing. This revamp has resulted in software becoming an integral part of the manufacturing – starting from the car design to the industrial design, to the electrical and mechanical design leading to the autonomous driving development capabilities. The paper begins by setting the context in the first section, discusses the issues in the second section and concludes with a few suggestions in the final section.

Table of Contents

Abstract

Introduction

The Business Rationale behind the development of AVs

Business Issues in AV development

IT infrastructure Issues for AV Development

Sensors

Context

The IT Infrastructure Discussion

Issues to consider in IT Infrastructure

Cybersecurity

Context

Cybersecurity in the context of Transportation as a Service (TaaS) model

Issues to consider in IT Infrastructure

Conclusion

References

 

 

 

 

 

 

Introduction

 

In this section we start by looking at the factors that contribute to the need of AVs and move to the corresponding business issues that face the industry. We conclude by deducing the current corresponding IT infrastructure issues from the business issues that are faced by the AV industry and predict a few that may arise in the future as the industry evolves.

The Business Rationale behind the development of AVs

 

The evolving technology of Autonomous Vehicles has far-reaching applications and implications that lie well beyond all current expectations. A major contributing factor to such a prospective AV development is the communication among cars and over infrastructure (connected vehicles). Looking at the next generation of AVs, the following motivations present a strong case for a move to autonomous vehicles, and thereby to a potentially enormous disruption to all aspects of life including business and society.

Reduction in road traffic casualties According to the WHO, road traffic deaths hit 1,25 million in 2013. Per the latest studies human error accounts for more than 90% of road fatalities. More than half road fatalities occur among cyclists, pedestrians and motorcyclists leaving high opportunities for improvement using autonomous driving technologies.

Reduction in social & environmental costs of driving Close to 9 billion people are predicted to live in urban areas in the next 25 years and automakers are under pressure to reduce the social and environmental impact of vehicles. AVs have the potential to facilitate vehicle sharing and ease traffic flow.

 

Potential improvements of access to mobility Once fully autonomous cars are available, significant improvements are expected in access to mobility. Thus, these technologies are likely to act as major enablers for the young, the (increasingly numerous) elderly and for those with physical limitations alike.

Economic benefits of making travel time productive Per Morgan Stanley, the driverless cars could contribute up to $1.3 trillion in annual savings to the US economy alone. Focusing on productivity increase, it lists the reasons for such benefits as the ease of travel and to work while traveling.

Business Issues in AV development

 

Mobileye, acquired by Intel in 2017, is a firm believer in the above reasons. The Israeli-based maker of self-driving tech has partnerships with more than two dozen carmakers and has become one of the leading players in the space. A recent interview of Amnon Shashua, the CEO of Mobileye is an excellent segue into understanding the business issues that the AV industry faces in light of the technology development to address the economic opportunity presented above. He spoke of the following challenges:

Safety. From a technical point of view, driverless technology can be split into two parts: its perception capability and its decision-making capability. The first challenge is to build a self-driving system that perceives the roads better than the best human driver. In the USA, the car fatality rate is approximately one death for 1 million hours of driving. Taking out drunk driving or texting instances, the rate almost decreases by a factor of 10. Effectively, this means the performance of a self-driving car’s perception system should not show a failure of more than, at an absolute maximum, once in 10 million hours of driving.

However, currently the best driving assistance systems show failure rates of once every tens of thousands of hours (Shashua, 2018). Other than improving the computer vision, he believes two other necessary components are required to close this performance gap. The first component addresses the creation of redundancies in the perception system using radar, cameras, and lidar. The second is to improve the quality of the detailed maps to make it easier for the vehicle to process the surroundings.

Usefulness. The second challenge for the AV industry is to build a system that makes reasonable decisions, such as when to change lanes and how fast to drive. Anytime a car without a driver makes a decision, it is making a trade-off between usefulness and safety. This is more of a regulatory issue, but the automakers must be able to program the vehicles to act within the formalized bounds of reasonable decision-making as decided by the regulators.

Affordability. The final challenge is to create a cost-effective vehicle, so that consumers become keen to switch to driverless. In the short term, because the technology is still at tens of thousands of dollars, only a taxi or ride-sharing business will be financially sustainable. This is so because removing the driver will potentially save the equivalent of tens of thousands. However, individual consumers aren’t likely to pay a premium of more than a few thousand dollars for the use of new technology. Thus, in the longer term, if auto manufacturers intend to market driverless passenger cars, they have to create much more precise systems than those that exist today at a fraction of the current cost. Thus, it is likely that robo-taxis come first while passenger cars follow within a couple of years.

IT infrastructure Issues for AV Development

 

For understanding the relevant technological challenges from the perspective of a CIO of an AV manufacturing company, it is imperative to understand the large number of cooperating functions that make the functioning of an AV possible and are central to the three strategic points made above. At a high level these functions include:

1. Interface for an end-user who needs to decide the goals for the vehicle such as destination, characteristics that define the route, intermediate stops if any and an ETA

2. A navigation system that is adept at planning an appropriate route and developing machine executable instructions to follow this route based on the vehicle’s position in real-time

3. A perception system which gauges the external environment’s situation and focuses on threats and safety related constraints such as objects, pedestrians, other vehicles, etc.)

4. A vehicle health system that records the vehicle state including fuel levels, parts and speed

5. Active vehicle safety system capable of using the available data from the perception system and the vehicle health system to ensure safe actions and plan movements

6. A vehicle control system which can take instructions from the navigation system (run left/right etc.…), the vehicle perception and health systems and issue control commands to the vehicle actuators with the permission of the active safety system

7. Actuators that translate the instructions into actions on the physical driving components such as steering, braking, etc.…)

8. Sensors that are the primary source of data to the various systems mentioned above

The complexity necessitated by the requirement of such functions leads to several very serious problems, notably the cost and time of integration of the systems, the challenges of testing this integration, assurance and data quality and above all, the difficulties of ensuring adequate performance and confidence throughout the life of a product which must be maintained and adapted over many years. We restrict our focus to Sensors, that are the primary sources of data, and Cybersecurity of this data that is critical for functional, legal and ethical working of all the cooperating functions mentioned above.

 

Sensors

Context

For close to a decade, relative positioning sensors for example radar sensors, have been used in commercial vehicles. The last few years have seen camera systems find their way into premium vehicles for abetting collision avoidance, in-vehicle traffic sign recognition and lane-keeping assistance. The first developed prototypes of fully-autonomous vehicles have made use of three-dimensional laser scanners in order to obtain an accurate representation of the surrounding environment.

However, passive or non-cooperative approaches, those in which a target vehicle does not actively contribute to the estimation of its relative position, have some fundamental issues. The most popular non-cooperative technologies that have been used over the years, such as radar, laser scanners and vision-based systems have a common problem – their line-of-sight characteristic. There are numerous situations during which these sensors are unable to offer a relative position and/or velocity estimate; the limited range, sight blockage by the surrounding topography or other nearby vehicles and the limited field of view are such common situations.

 Furthermore, even the cooperative transponder-based approaches that use RSS, RTD and TOA measurements exhibit large errors in non-line-of-sight conditions. This is considered an important drawback for vehicular safety applications that need to react in a timely manner to dynamic events ahead of the vehicle immediately in front. For this reason, cooperative approaches based on Vehicle to Vehicle(V2V) communication, which can cope with occlusions of the line-of-sight up to several hundreds of meters, offer a real advantage.

The fusion of both cooperative and non-cooperative approaches yields the most promising relative position estimation performance. It is suggested to combine the high accuracy and good robustness against the lighting and climatological conditions of radar sensors, with the extended all-around range and identification capabilities of V2V communication. Vision-based systems and radar sensors could in the future incorporate the information on surrounding road users provided by a cooperative technology at the lowest-level to improve vehicle detection and the resolution of different targets. For a cooperative approach, it is envisioned that the exchange of GNSS-derived pseudo range and carrier-phase measurements for differential positioning along with kinematic and inertial sensor information will provide the highest accuracy, availability and robustness.

The IT Infrastructure Discussion

 

That fusion is the best way ahead can be deduced from the comments made by Bryan Salesky, CEO Argo AI, a Startup promoted by Ford Motors that is working on AVs. He states that LiDAR sensors work well in poor lighting resulting in the grab of surrounding three-dimensional geometry, but don’t provide texture or color; they have to use cameras for that. On the other hand, cameras are dysfunctional in poor lighting. Contrastingly, radar which is relatively low resolution, can directly detect the velocity of road users even if they happen to be at long distances. This is the reason for so many sensors still being mounted on any autonomous car — the strengths of one type complements the weaknesses of another. Also, individual sensors are unable to reproduce what they capture most of the times, so the computer has to corroborate the inputs using data from multiple sensors, and then sort out the inconsistencies and errors.

While combining all of this into one comprehensive and robust picture of the world for the computer to process is incredibly difficult, developing a system that can be manufactured and deployed at scale with cost-effective, maintainable hardware is even more challenging. Focussing on innovation across the software stack and sensing hardware to lower costs has to be an organizational goal.

Issues to consider in IT Infrastructure

 

There remains significant work to be done to accomplish these conflicting objectives and get the technology to reliably scale. The CIO must start by aligning the following key variables with the government regulations and company strategy.

Accuracy

Generally, the accuracy of any measurement system can be defined as “how close” the measurement is to its accurate value. This closeness is supposed to be quantified by the distance between the measured and correct value. In most positioning systems, accuracy is provided in three dimensions.

Reliability

For applications that are safety-critical, the system’s reliability is an extremely important factor. Even if the accuracy of a given system is very high, it cannot be banked upon if it is not adequately reliable. While all operational components involved in any civil aviation transport adhere to strict requirements, such imposition is a new concept in road transport.

Availability

That any positioning system is readily available when called upon most of the time is of high importance. There can be multiple reasons that may cause diminished availability of a system. As an example, a positioning system based on GNSS may not be available for use in situations when there is absolute obstruction in the line-of-sight to the satellites, such as in tunnels. 

 

Detecting Field of View and Range

A few ranging systems, such as visual systems or laser scanners, exhibit line-of-sight characteristics, which means that they are able to measure only the position of those neighbors to whom they may have direct visibility.

Dimension

Any position in space is a 3-D component, meaning it is defined as a relative position coordinate. However, a lot of relative positioning systems are unable to measure relative positions in the three-dimensional space. 

Target Resolution & Identification

Resolution of targets may be defined as the system’s ability to resolve different objects. This is an important component of driver assistance systems because it helps the system to quantify the actual number of targets and ensure that these are tracked over time.

System Delay

For driver assistance applications that are relevant for safety, a fast and dynamic response is required. This characteristic is central to the warning system’s ability to administer timely alarms. The same also helps the controlling system to respond smoothly to any changes that might have occurred in the AV’s relative position.

 

Non-Technical Requirements

There are other limitations or non-technical requirements that must be addressed when evaluating the suitability of any relative positioning system. For example, the price plays an important role in freight and commercial passenger vehicles. The cost component of any relative positioning solution also includes installation and maintenance costs. As well, costs such as those associated with maintaining the required processing power, power consumption, and maintenance of noise and heat creation. In solutions that make use of infrastructure-based communications, such as cellular communication, operating costs that may be monthly or annual fees can be reasonably expected.

While research and empirical evidence points towards the fact that road vehicles can rely solely on the on-board perception sensors, it is now reliably being forecasted that the same vehicles will greatly benefit from the introduction of Vehicle to Vehicle communication. This is so because not only does V2V communication abet an increased reliability and availability in cooperative relative positioning, it also enables the cooperative perception because of its ability to share sensor information and information about the execution of collaborative movements between other automated vehicles on the road. Thus, a higher degree of safety is likely achievable without sacrificing any efficiency.

 

Cybersecurity

Context

In a candid interview, Nissan’s CIO Celso Guiotoko talks about how at Nissan everything inside the autonomous vehicle has been made the responsibility of IT; interestingly, everything outside of the vehicle has been made the responsibility of IT too. This means that any device from the vehicle that sends data to the cloud, whether from apps or web pages, is now being developed collaboratively by the IT department and the engineering team. He backs this approach because he doesn’t understand how the testing of various components can be fruitful in the end if the teams keep working in their own silos. Man-management skills to ensure collaboration of this sort will go a long way in helping the CIO deliver upon the successful technology component of the AV.

Guiotoko also goes on to single out cybersecurity as the top concern for him in the group’s quest for the development of its autonomous line of vehicles. Issues to manage for the CIOs include ensuring the profitable use of the enormous amount of data that the cars will gather in the primary process of driving; while it is not yet clear what the business model will become in the future specially in terms of the monetization of the data stream, there definitely is clarity about the intent to collect data at scale and use it to increase the turnover of the company.

Increasingly, auto manufacturers are making cars that are connected to each other – it’s the Internet of Things (IoT) restricted to cars for the moment but with the potential to connect to other devices later. As more consumers drive more such cars, the malicious actors have a bigger target to aim at; this is much more dangerous than a credit card mishap because a hacker who gets into a vehicle’s connected system may even cause the driver or travelling passengers to lose their lives.

Cybersecurity in the context of Transportation as a Service (TaaS) model

People all over the world spend billions of hours in their vehicles every year. Also, this number is only set to go up. With the advent of TaaS as a business model, several carmakers have started manufacturing their autonomous vehicles for not only transportation but also to ensure that the vehicles provide adequate tools for a significant increase in work productivity on the fly. This initiative is set to free up the time that is otherwise unproductively spent driving. This free time is supposed to have a huge impact on productivity by increasing the productive time spent on activities such as sleeping, working, studying, etc.

Volkswagen level 5 autonomous concept car has an AI powered ability to configure the car’s interior space for different activities including work. According to Dr. Dieter Zetsche, Chairman of the Board of Management of Daimler AG and Head of Mercedes-Benz Cars, cars of today perform various roles in addition to their traditional role as a means of transport. He goes onto verbalize the German carmaker’s vision as a manufacturer of such cars that will eventually have the ability to transform into a mobile living space.

The flip side of this concept of smart mobility is that it presents CIOs with additional headaches. Managing the security of the vehicle navigation system is one thing while ensuring the infrastructure and security for smart in-car connectivity is another. Sure, not all autonomous vehicles will be solely designed to provide the additional capability to serve as an office on the go, but there is a distinct possibility that people in the major cities will, in the future, need vehicles that are equipped with the tools which can provide the functionality of a moving office. And that’s not where it stops; autonomous cars are also on their way to becoming the very core to the Internet of Things (IoT), a fact that presents another significant consideration for CIOs in their quest of handling the information security of this extended office. Thus, AV Cybersecurity risks are sure to be amplified multifold in light of the growing demand of the TaaS model and the role of the AV as a core element of the IoT.

Issues to consider in IT Infrastructure

 

Thus, CIOs need to ensure that they prepare accordingly and that the right pieces are in place. This is not something to be done over a short period of time, but it is the reality the AV industry faces in the not too distant future. To that end, the CIOs have to look ahead and meticulously create very clear guidelines in terms of how they will ensure the security of the connection, the vehicles and how they make sure that only drivers and customers have access to the autonomous vehicle. The following issues must be addressed by the CIO in the security infrastructure of AVs:

Security by Design Secure vehicle design involves the integration of hardware and software cybersecurity features during the product development process.

 

Risk Assessment and Management Risk assessment and management strategies mitigate the potential impact of cybersecurity vulnerabilities. Best Practices focus on processes for identifying, categorizing, prioritizing, and treating cybersecurity risks that could lead to safety and data security issues. Risk management processes help auto manufacturers identify and protect critical assets, provide support for operational risk decisions and provide assistance in the development of protective measures. 

Threat Detection and ProtectionProactive cybersecurity through the detection of threats, vulnerabilities, and incidents empowers automakers to mitigate associated risk and consequences. Threat detection processes help raise awareness of suspicious activities by enabling proactive remediation and initiating recovery activities.

Collaboration and Engagement with Appropriate Third Parties Defending against cyber-attacks often requires collaboration among multiple stakeholders to enhance cyber threat awareness and cyber-attack response. When faced with cybersecurity challenges, the industry is committed to engaging with third parties, including peer organizations, suppliers, cybersecurity researchers, government agencies, and Auto-ISAC, as appropriate.

Incident Response and Recovery An incident response plan documents processes to inform a response to cybersecurity incidents affecting the motor vehicle ecosystem. Best Practices include protocols for recovering from cybersecurity incidents in a reliable and expeditious manner, and ways to ensure continuous process improvement. 

Governance The purpose of aneffective governance program is to align a vehicle cybersecurity system with the organization’s broader mission and goals. Furthermore, strong governance can help to foster and sustain a culture of cybersecurity. Best Practices do not dictate a particular model of vehicle cybersecurity governance but provide considerations for organizational design to align functional roles and responsibilities.

 

Awareness and Training Training and awareness programs help cultivate a culture of security and enforce vehicle cybersecurity responsibilities. The Best Practices emphasize training and awareness programs throughout an organization to strengthen stakeholders’ understanding of cybersecurity risks.

Conclusion

 

In the face of the issues discussed till now, the traditional IT architecture suffers from the following shortcomings: –

•         Unstructured and inconsistent approaches to such things as error management, thread priority management and vehicle to vehicle or vehicle to network communications

•         Inefficient approache to handling peak CPU usage

•         Unstructured and inconsistent data and software interaction semantics

•         Low bandwidth networking

•         Divergence in timing

•         A lack of modularity at the software level

•         Inflexibility in system re-engineering and re-configuration

Addressing these problems requires a new architectural approach with the goal of satisfying the traditional needs of performance with direct attack on the problems raised by complexity. Future architectures should consider:

•         A rethink of the HW structure towards a more centralised compute resource with more power available to be managed for peaks across functions. This approach may also lead to a significant reduction in weight and cabling complexity within the AV

•         The introduction of networking that is mainstream IP based, albeit with provision for some specific extensions where required. The provision will serve to support mainstream development styles as well as the need to add bandwidth.

•         The initiation of a flat data plane for access to all data, including for Audio and Video. This move will help the acquisition of data by software in a near absolute manner significantly reducing configuration issues.

•         Adherence to a stringent definition of admissibility for software elements. This will result in certain aspects of software behaviour to be uniquely defined at the levels of syntax and semantics for all the software. For example, in software lifecycle a unique API could be enforced for all software elements.

•         A support level enforcement for location transparency across software transactions. This will result in all the software to interface an underlying and transparent communication layer. Thus, the software won’t require to know the system topology for the purpose of communication.

•         Direct architecture support with respect to security structures. This allows security to be managed as a configuration activity during system integration and based on standard system elements.

Summarily, software is becoming an integral part of the manufacturing – starting from the car design to the industrial design, to the electrical and mechanical design leading to the autonomous driving development capabilities. The IT architecture will have to comprise of hardware and software that presents a unified feature platform for the entire AV, not just a standalone ECU. This unified architecture will extend from the data center to the vehicle. It can be trained at a central facility, tested and validated in simulation and deployed in the vehicle’s AI. This change would support a more targeted approach to system engineering while ensuring that key processes are in place to tackle the growing complexities of the evolving system while trying to ensure safety and cybersecurity at all steps.

 

 

 

References

Arslan, S. (n.d.). The Future of Autonomous Transportation requires AI supercomputing in the Car and the Cloud. Retrieved July 15, 2019, from https://automotive.cioapplicationseurope.com/cxoinsights/the-future-of-autonomous-transportation-requires-ai-supercomputing-in-the-car-and-in-the-cloud-nid-841.html

Autonomous Vehicles: Cybersecurity is The Key Factor. (2019, February 19). Retrieved July 15, 2019, from https://www.cioapplications.com/news/autonomous-vehicles-cybersecurity-is-the-key-factor-nid-3464.html

Bagloee, S. A., Tavana, M., Asadi, M., & Oliver, T. (2016, August 29). Autonomous vehicles: Challenges, opportunities, and future implications for transportation policies. Retrieved July 15, 2019, from https://link.springer.com/article/10.1007/s40534-016-0117-3

Best Practices – Auto-ISAC. (2016, July). Retrieved July 15, 2019, from https://www.automotiveisac.com/best-practices/

U.S.Cong. (2018). Issues in autonomous vehicle deployment(B. Canis, Author) [Cong. Rept.].

Hao, K., & Hao, K. (2019, April 23). The three challenges keeping cars from being fully autonomous. Retrieved July 15, 2019, from https://www.technologyreview.com/s/613399/the-three-challenges-keeping-cars-from-being-fully-autonomous/

Harshman, J. (2018, May 16). Nissan autonomous vehicle: The man keeping hackers at bay. Retrieved July 15, 2019, from https://www.hottopics.ht/21876/nissan-autonomous-vehicle-the-man-keeping-hackers-at-bay/

Jackson, D. (2016). Autonomous Driving : How to overcome the 5 main technology challenges?(Working paper). Altran.

Müller, F. D. (2017, January 31). Survey on Ranging Sensors and Cooperative Techniques for Relative Positioning of Vehicles. Retrieved July 15, 2019, from https://www.mdpi.com/1424-8220/17/2/271/htm

Salesky, B. (2017, October 21). A Decade after DARPA: Our View on the State of the Art in Self-Driving Cars. Retrieved July 15, 2019, from https://medium.com/self-driven/a-decade-after-darpa-our-view-on-the-state-of-the-art-in-self-driving-cars-3e8698e6afe8

Sethi, K. (n.d.). Why CIOs Should Take Autonomous Vehicles and TaaS More Seriously. Retrieved July 15, 2019, from https://www.readitquik.com/articles/digital-transformation/why-cios-should-take-autonomous-vehicles-and-taas-more-seriously/#

Vining, J. (2019, June 09). Autonomous Vehicles: What Transportation CIO’s Should Consider. Retrieved July 15, 2019, from https://www.thansyn.com/autonomous-vehicles-what-transportation-cios-should-consider/