First people thought automated driving was a sprint race, now many think it’s a marathon. However, it’s turning out to be an ultra-marathon relay. No one can solve its challenges alone: the future of automation lies in collaboration.
Safety trumps accelerated deployment
Leading up to 2018 tech news was full of claims of fully self-driving fleets and widespread adoption by 2017, 2018 and later 2021. This optimism is now nowhere to be seen. A series of accidents sadly leading to fatalities served as a wakeup call to the largely unregulated automated driving industry that the approach had to change. The focus shifted from realizing autonomy as quickly as possible to making it as safe as possible in the long term.
In practice, this has resulted in OEMs and Tier1s already reorganizing their automated driving research and development teams. Car manufacturers are now looking into deploying existing technologies as advanced driver assistance systems (ADAS). Only a select few stakeholders are still aiming to control the whole development process and deploy their own fleets, the wider industry meanwhile, is increasingly open to collaboration.
Collaboration feeds standardization
The reason behind this is simple: collaboration is the key to survival. Automated driving is immensely complex and resource intensive. No one will be able to solve every problem alone. However, the approach to the development of these systems has to change to make collaboration viable. Software solutions must increasingly move towards hardware agnostic modular designs to facilitate deployment, while industry standards are needed to enable knowledge and data exchange.
Understanding the inner workings of Artificial Intelligence-based networks used for object detection serves as a good example of the challenges ahead. As opposed to classic algorithms, Artificial Intelligence-based networks are largely shaped by training data and behave in a nonlinear way. This can lead to results not necessarily understood by the network designer or the end users. The automotive industry, however, has low tolerance for uncertainty in how accurately automated cars can detect and classify objects in their way. To move forward with the widespread deployment of automotive AI, the industry will need to agree on a framework to provide methods and tools for the assessment of these networks.
Standardization will either be achieved through cooperation and self-regulation or forced upon the industry by regulator and governments. An example of such would be enforcing compulsory “self-driving tests” for automated vehicles. The grounding of Boeing 737 Max 8s globally following the tragic events of recent months is an example of how drastically regulators can limit new technologies if their respective industries fail to regulate and standardize themselves. Naturally, the automated driving sector must strive to avoid such difficulties and work towards making the testing and deployment of automated driving as safe as possible.
Add in decreasing investor momentum in the automated driving industry, the slowing global economy and the shrinking automotive market, and the need for building meaningful collaborations and working towards trusted industry standards becomes increasingly evident. I look forward to discussing automotive industry collaboration to establish an AI framework at the forum session on “AI for ALL? Building on a global ecosystem to boost societal and economic potentials” at ITU Telecom World 2019 in Budapest this September.