In early 2019, Elon Musk claimed loud and clear that a self-driving Tesla would be so autonomous by the end of 2020 that you would be able to sleep at the wheel. Yet others working with autonomous transportation are far from sharing the same optimism. The premise of autonomous vehicles promises lower production and maintenance costs, improved safety and more efficiency. Technology enabling autonomy is already well under development, yet the expected time of deployment varies across industries, and relies on public trust and national and international regulation.The anatomy of autonomy
The level of autonomy, and consequently human supervision, ranges from simple driver assistance to fully autonomous operations. Based on the autonomy level and purpose of transportation, there are different requirements for technologies such as sensors, analytics, algorithms and actuators. Be it a car, train, ship or an aerial craft, for a vehicle to be able to operate with less human involvement the same considerations must be carried out1:
- Condition detection:
An autonomous vehicle will need to be able to sense information from the environment in which it operates. This sensing can be carried out by a multitude of sensors such as lidars, radars, cameras, sound detectors as well as satellite information and other communication networks. Sensorial information that must be collected ranges from objects in the vehicle's vicinity, environmental conditions, such as weather and friction, to the vehicle itself such as engine condition. The quality and reliability of the sensors are of crucial importance.
- Condition analysis:
Once sensor input is gathered, the vehicle can use the fused sensorial input in combination with algorithms and advanced analytics to assess the current condition of the vehicle and its surroundings. The algorithms categorizing these inputs will need to further predict the future states of collected information and identify potential risky situations.
- Action planning:
Once the condition of the vehicle and its surroundings is established and analyzed, the course of action must be planned. This is done by an algorithm, either a self-learning algorithm or a rule-based pre-programmed one, or a combination of both. Rules and regulations for safe operation, such as the International Regulations for Preventing Collisions at Sea (COLREGs) or common traffic rules, must also be incorporated and adhered to. A self-learning algorithm can be trained from data generated by a simulated environment, by field testing or by operational data. Most likely it will be a combination of these. Using simulated and augmented data will most likely be necessary to ensure that diverse enough and sufficient number of scenarios are covered.
- Action control:
When the action has been planned and a decision has been made, control systems and actuators make sure that the decision is carried out by ascertaining appropriate steering and machinery output. For an autonomous system, the control commands will be generated and sent from the action planning software to the control system. The reliability of the action control will depend on the reliability of the control system and the actuators.
It is envisaged that ships will be the primary mode of transport that will require remote control. It is, however, likely that remotely operated ships must be equipped with a certain level of fully autonomous capability as a fail-to-safe mechanism. In most autonomous ship concepts, there is a combination of remote control and automation, where each of the above steps can be performed either manually (local or remote), by automation or by a combination of these. For cars, trains and planes, fully autonomous operation is more likely.Proceeding with caution
With the expectation that introducing autonomy into transport would achieve significant cost and efficiency gains on a global scale, the level of optimism and investment in autonomous vehicles’ early years isn’t surprising. However, there has been a notable recent shift, from the all-in attitude previously communicated by vehicle manufacturers to one of stepping forward with caution.
With decreased cost of sensors, data storage and processing power, the costs of autonomous systems are likely to decrease and these systems, in theory, operate safer than their human counterparts. Autonomous vehicles hold great promise in terms of improved safety performance as many accidents in the transport sector can be attributed to human error. By one account, driverless transport will add $7 trillion to the global economy and save hundreds of thousands of lives in the next few decades2. Autonomous vehicles will also enable more efficient supervision and handling of traffic and logistics, with vehicles seamlessly communicating with each other and other parts of the logistics chain. A more efficient transport sector will in turn lead to less CO2 emissions, both because vehicles can be better utilized and have more space for payload, but also because the energy requirement will be potentially lower.
However, public acceptance of autonomous vehicles and scaling of the technology relies on the safe and reliable introduction of this technology, whilst at the same time ensuring that the technology is both cost-effective and regulated – no mean feat when operating on cross-national grounds. Regulation, requirements and verification of compliance will differ for different modes of transportation, and autonomous vehicles will not scale before this is in place.Taking the wheel back on driverless cars
When the driverless car is mentioned, Tesla tends to spring to mind. Currently, all new Teslas are claimed to have the hardware needed to carry out self-driving operations in most conditions, albeit under human supervision, and the technology will develop to be completely autonomous within 2020 – if Elon Musk’s ambitions are rooted in cause. Yet most major car manufacturers have programs to develop self-driving cars, so why are there not more frontrunners?
John Krafcik, CEO of Google’s self-driving car project Waymo, believes there will always be restraints on autonomous vehicles that restrict them for performing to full capability3. Meanwhile, Toyota has been comparably slow in researching autonomous technology and Nissan abandoned its plans to have driverless cars on the streets in 20204. A two-fold hurdle stands in the way: significant uptake of autonomous vehicles of any kind will not be seen before society perceives them as safe. This will require requirements, regulations and verification of compliance from independent bodies. Moreover, verification of autonomous vehicles and AI is unconventional and will therefore require new tools and methodologies. As it currently stands, public trust in driverless cars is shaky and trans-national regulation has not yet materialised. On the existing track, it is likely that fully self-driving cars will not be commonplace before 20255.Unmanned voyages to sea
The application of autonomous technology to shipping has been under debate for some time: both for the environmental and economic advantages it could provide, and for the challenges in regulation. But in short, it’s unlikely that there will be an industry-wide adaptation of autonomous ships before 2030.
Several projects with the aim of introducing and taking steps towards unmanned ship are already underway6 with a notable example being the Yara Birkeland project which aims to move to fully autonomous operations by 2022. However, these are typically specialized projects with a limited operational area that either fall under national legislation or are excepted from international legislation. The general development is slow, and it is expected that navigational assistance and remote-assistance would be the first steps within the maritime industry.
Although interim guidelines for trials of autonomous ships7 exist, regulation of autonomous vehicles in the shipping industry requires rigorous amendments to existing regulations and the creation of new regulations. This is especially relevant for international rules related to safe manning and navigation.The road ahead
To sum up, Tesla claims to have completely autonomous cars ready within 2020; such vehicles are not expected to be commonplace before 2025; unmanned shipping operations could begin by 2022; and industry-wide adaptation of autonomous ships could be expected within the decade but not before towards the end.
Autonomous transport at any level cannot have market traction before regulation and public trust are in place. Driverless vehicles must be deemed as safe, or safer than, human-operated vehicles and this perception relies heavily on the state of regulation. Adequate regulation must consider the robustness of AI to carry out safety critical assessments, which as of now is still unclear. Assurance frameworks for digital assets and autonomous vehicles must also be developed.
Regulation must also cover cyber threats, which increase with a vehicle’s reliance on software and connectivity. This threat is even more evident for autonomous vehicles where all the operational functions are controlled by software and control signals sent through a communications link.
Further still, with whom the liability lies in case of damage caused by an autonomous vehicle to property or humans is still under development in law and policy. This will require an evolvement from current liability laws.
The cost incentive of moving from manned to unmanned vehicles will also have a part to play in the rate of market adoption, as will the perception of negative or positive impact on society. Although new expertise will be needed in the industry of driverless transport, the loss of more traditional transportation jobs is a risk. It is also likely that this change will lead to an unequal impact on different societal groups and regions, as the competence and location of these workers will be different from what is required today.Contributors
Main author: Hans Anton Tvete
Editor: Tiffany Hildre
- DNV GL, 2018, Position Paper; Remote-controlled and autonomous ships