Is There a Need for Remote Control for Autonomous Vehicles?

Original Equipment Manufacturers (OEMs) are implementing ambitious deadlines for highly automated driving in the next three to four years. Much of the focus of automotive technology development has turned to addressing implementation pain points such as consumer adoption and maintenance rather than the core fundamentals. Indeed, the core-enabling technologies for autonomous driving are well established, with broad consensus on the need for multiple diverse sensors, high-compute platforms, high-bandwidth/low-latency networking, and artificial intelligence.

A relatively recent innovation for the automotive industry is the use of simulation to self-certify that an autonomous system is robust through targeted testing and training of neural networks. Nevertheless, it is broadly accepted that autonomous vehicles will not be able to navigate every situation they encounter—certainly not during the early years of adoption.

To satisfy functional safety requirements, autonomous system developers can specify last-resort, minimum-risk maneuvers which bring the vehicle to a safe state; this usually means gradually slowing the vehicle to a halt in a safe space. While this may help OEMs prevent dangerous situations for which they may be liable, questions remain as to how consumers will react to an autonomous vehicle suddenly coming to a halt. In semi-autonomous vehicles, it would be relatively simple—if slightly frustrating—for the driver to reassume control and navigate the vehicle.

In fully driverless vehicles, however, the vehicle occupants will have no means of taking over the driving. This has led to a number of players suggesting remote control or teleoperation as a remedy. However, it is important to question the market potential for such a niche application and, more broadly, consider what long-term future there is for technologies which are targeted at frictions associated with the early years of implementation.

The market potential for remote control is heavily tied to that of the Society of Automotive Engineers (SAE) Level 5 robotaxi implementations and the rollout of 5G networks. High-bandwidth and low-latency connections will be essential to safe operation. In order for remote operators to pilot the vehicle successfully, they will need to perceive the vehicle’s surrounding environment as well as a physically present driver would and rely on flawless High-Definition (HD) video streams to gain an accurate and timely understanding of the traffic situation. In addition, the driving inputs (steering, acceleration, deceleration, etc.) would need to be sent with ultralow latency so that the remote driver’s visual feedback and operational inputs are harmonized. Finally, as with all mission critical automotive functions, security and reliability are paramount. Only 5G, with support for network slicing, can offer safe remote control.

As discussed above, remote operation will only be relevant to fully autonomous implementations; expecting the consumer to abandon the vehicle and continue on foot is obviously unacceptable.

The ecosystem for remote-control operation is therefore still quite limited:

• Huawei, China Mobile, SAIC Motor: An iGS concept car was operated by remote control as part of a 5G demonstration. The remote driving test saw a driver maneuver a vehicle from 30 kilometers (km) away, relying on numerous HD video feeds broadcast in real time to provide a 240° appreciation of the vehicle environment. End-to-End (E2E) latency for vehicle control functions was such that when the vehicle was traveling at 30 kilometers per hour (km/h), the distance traveled between the initiation of a braking command and deceleration was only 8 centimeters (cm).
• Nissan: Nissan has set out their vision for a Seamless Autonomous Mobility (SAM) system, whereby struggling autonomous vehicles will come to a halt and await intervention from an operator in a remote command center.
• NVIDIA: At the 2018 NVIDIA GTC Conference in San Jose, the opening keynote featured a live demonstration of remote operation using a Virtual Reality (VR) headset.
• Phantom Auto: A dedicated startup in the remote-control space, Phantom Auto offers remote-control operation as a service, providing an Application Programming Interface (API), software for high-bandwidth/low-latency communications, and a team of trained operators.

ABI Research expects that by 2030 there will be 11.4 million SAE Level 5 vehicles on the road, all deployed in robotaxi applications. In terms of penetration, this will amount to around only 1% of all registered vehicles. From this perspective, the market opportunity for any initial implementation support seems relatively limited. That said, it is important to consider the unique way in which SAE Level 5 will spread. While traditional automotive technologies have spread through a conventional trickle-down approach (i.e., being adopted at high cost on premium vehicles before achieving economies of scale and spreading to more mass market vehicle segments), the expansion of Level 5 will be geographical. That is, smart mobility operators will deploy the vehicles in a geo-fenced area, gradually expanding that area as regulations, maps, and supporting infrastructure are rolled out. Therefore, technologies targeted at easing some of the pain points of implementation will maintain their relevance for years to come, as robotaxi service operators expand into new cities and new countries. As the level of experience in each new geography grows, the need for services like simulation and remote control will diminish until the next expansion occurs.

While some OEMs are looking to develop remote control as an in-house competence (or are in the very early phases of doing so), the niche size of the market at any given time suggests it is best suited to a specialist third party who can offer the capability as a service. At the same time, more long-term and sustainable use cases should be identified and exploited, with remote operation of specialist vehicles in hazardous environments being a prime target.