Autonomous Dream Could Face a Data Logjam

As vehicles become increasingly connected, different stakeholders are looking to benefit from the information that connected and autonomous vehicles (CAVs) provide.

However, as TU-Automotive contributor Nathan Eddy says, the key challenge is to ensure that the data is being sent to its intended location without delay. Data from or to the vehicles can be hampered by latency and packet loss. With CAVs evolving there will be increasing data volumes, putting pressure on the networks. At present CAVs may be able to cope but will they be able to transmit and receive data in a sufficiently timely manner given that the networks can be slowed down significantly?

Permissible limits

Suhas Gurumurthy, industry analyst connected cars at Frost and Sullivan, says his company recently conducted some “extensive research on data related to connected and autonomous vehicles”, and it has found that there some industry contacts consider the latency and packet loss issues as being “within permissible limits as long as they are not safety-critical applications”.

“For example, if it is a video feed for autonomous vehicle algorithms to interpret, then there is no room for packet loss and latency. Any form of distortion in the video quality and a slight delay in receiving the data packets will be a recipe for disaster. In such applications, high-end networking, low-latency technologies with high packet delivery rate is essential. 5G will be able to achieve this by bringing the latency to about 1ms thereby reducing packet loss.”

However, latency and packet loss become more of an irritation than a safety issue when it affects the streaming of a song via Apple Music or Spotify, or when it affects some other form of entertainment or infotainment. The most severe consequence could be a loss of brand loyalty rather than an injury or a death.

Basic safety messaging

Thankfully, he reports that there have been some improvements in the delivery of basic safety messages (BSM) in vehicle-to-everything (V2X) scenarios. He explains: “Latency is always cited as a concern, and drastic improvements have been achieved both in terms of DSRC and C-V2X technologies. In the most safety-critical use-cases can now achieve 50 to 100ms from what used to be in seconds.”

His Frost and Sullivan colleague, Manish Menon, team lead chassis, safety and autonomous driving, mobility, adds there are a range of use cases, including for remote tele-operations, whereby remote servers are using to control CAVs remotely. Remote operations require high bandwidth, and it is crucial to avoid having latency and packet loss issues to avoid losing remote control of the vehicles.

“When it comes to safety, latency and packet loss can lead to accidents as well as injuries and loss of life. The timeline, currently, we are seeing safety critical V2X (Day 1 use cases) being in place by 2024 or 2025,” says Gurumurthy.

Potential solutions

So, what solutions are available to CAVs to enable them to mitigate the effects of latency and packet loss?  To explain what is available, David Trossell – CEO and CTO of Bridgeworks, says it is important to “break down what the systems are likely to be in order to ascertain the solutions available, and their likely technologies”. He provides some potential options in the table below.

Solution	Technology
CAV to CAV communications	TCP/IP Mobile Data over 5G
On board camera	Optical Sensor derived decision making algorithms.
Traffic Lights/ Intersection controls	Backhauled TCP/IP stream from Traffic sensors to Data Center/ Cloud.
Security	IPSec/Diffie Helman Key exchange/ Certificates.
Collision Avoidance	Multiple from all of those above.

 

Trossell adds: “With the emergence of Big Data in the cloud, IPv6 and the internet of things we can ascertain that it is highly likely that a TCP based system will be key to CAV-CAV communications, intersection controls, collision avoidance and security. Certainly, User Datagram Protocol (UDP) is not a reliable mechanism and emerging protocols stacks are going to have to go a long way to replace the traditional Syn, Syn-Ack, Ack handshake.”

“TCP’s sliding-window and its reliability measures along with reliable session establishment will go a long way in achieving the communications goals of the CAV’s but it’s not a complete solution. client or server communications almost never communicate to their full potential unless latency and packet loss migration technologies are utilized.”

Resilience and channel bonding

He comments the possible mitigations of latency and packet loss lie in resilience and channel bonding throughout the network to ensure that multiple paths exist. There is also the hope that 5G will lead to network improvements, leading single digit latency in the sub-10ms to sub-1ms range. The specific latency rate will depend on the connected device application.

Further to these potential ways of mitigating latency and packet loss, he adds: “Fully optimized data center to data center WAN solutions. Live traffic data will be ‘Big Data’ and this live traffic data sent between data centers will need to be optimized or accelerated with traditional latency mitigation techniques at each location to ensure it is processed both speedily and reliably.”

Each of these ‘solutions’ will only so far to mitigating latency and packet loss, and so he emphasizes there will need to be some mandatory failsafe measures in place for safety reasons: “Vehicle manufacturers will have to improve on optical camera technologies and automatic braking technologies in order to provide ‘belt and braces’ measures. This will ensure tailgating and running through red lights are avoided. The system will also need to know to apply the brakes in a predictable manner, should something fail.”

System segmentations

Gurumurthy adds that latency and packet loss can be managed with system segmentations, which have “different modes of data transmission will lead to better flow and help in reducing latency and packet loss.” For autonomous driving safety-critical applications there are also multiple redundancies. He finds that the sensor suite “can take care of the OEDR (object and event detection and response) but SAE level 3+ applications would require validation of data through V2X and/or map data, and several other sources.”

Traffic sign recognition (TSR), camera sensors can detect signs. However, he explains, to “validate the detection requires map data in cases where the sensor data is unable to recognize the actual sign – whether that be a stop sign or something else, due to the system inefficiencies, packet loss, it may detect it as something else).”

Reducing cloud reliance

To reduce the transmission of data to the cloud, there is edge computing. “Edge computing could play a crucial role in safety critical use cases by taking up the processing on-board the vehicle, when there is a loss of network or if there is latency,” he claims.

To improve the efficiency of the networks, and to reduce the data logjam, it is argued that there is also a need for automakers to unify connectivity and autonomy over the next 4 years. “Currently the development of connected and autonomous vehicles is happening in silos, different teams, different technologies and no unified roadmaps which does not allow a unified strategy for data and also deployment of services,” reveals Gurumurthy.

Data ingestion and prioritization

For now, the current focus is on data ingestion, the prioritization of data sets, along with data orchestration, movement and compute for dynamic access needs. He believes there is a need for a unified development approach towards connected and automotive vehicles. OEMs are, for example, addressing this by deploying advanced connected services for storage and processing and service deployment, and OEMs are tackling test, development and the deployment of CAVs separately with multi-cloud strategies.

With autonomous vehicles, there is the prioritization of data collection too – involving between 3-10 TB per car per day. However, he comments: “It is unclear how much of data will be transmitted from deployed automated vehicles; it’s likely there will be an increase in data sent from the car.” The data focus for connected vehicles is on embedded over-the-air (OTA) updates for mostly, map, navigation, media, operator manual, and some of them also provide firmware upgrades.

Data bandwidth concerns

“When the entire OEM vehicle line-up starts pushing firmware upgrades, the data bandwidth would be a concern”, he adds. Meanwhile, data costs can be kept in check with smart sensors and with sensor fusion. There is also a prerequisite to enforce stringent cyber-security at both an organizational as well as at a vehicle level. Advanced artificial intelligence and machine learning can play a role too, helping to optimize the use of data for “for developing future services, both internally and externally.”