Autonomous driving currently available to the everyday consumer includes park assist, automatic braking, cruise and speed control and the like. As we get closer to fully automated vehicles, we can expect to see smart cloud computing software, blind-spotting tools and advanced cameras.
What is self-driving technology?
Self-driving technology, also known as autonomous vehicles and driverless cars makes use of artificial intelligence, cameras, radar and sensors that enable a vehicle to operate without a driver. Autonomous cars must be able to control and navigate safely.
Levels of automation
To be considered fully automated, a vehicle must graduate through five different levels:
- Driver assistance
- Partial automation
- Conditional automation
- High automation
- Full automation
Many models have successfully achieved up to conditional automation, showcasing how much further we have to go before we can expect fully autonomous vehicles to be readily available for purchase.
What is needed for driverless cars to become a reality?
Even if fully autonomous vehicles were readily available for consumer markets, there are additional issues that must be addressed to ensure these vehicles are safe to operate.
Poor road conditions
In order for driverless cars to operate fully, roads need to be upgraded. This includes filling in potholes and improving lane markings. Road conditions as they are will not be able to support high tech self-driving cars.
For self-driving cars to operate, they must hold the ability to connect with other autonomous vehicles. Volvo has developed a technology that enables their driverless cars to communicate with each other as well as highlight potential hazards on the road via a cloud-based network.
In addition to autonomous vehicles communicating with one another, they must be able to operate safely among non-autonomous vehicles.
Laws will need to be implemented outlining what weather conditions will be safe for self-driving cars and other vehicles to operate in. Additionally, it will be imperative to understand where liability lies if an accident occurs involving an autonomous vehicle.
In March 2020, Tesla filed a patent surrounding sourcing self-driving training data from its vehicle fleet as a way of training their self-driving neural network. Unlike other companies who are investing in self-driving functionality, Tesla relies on a more extensive fleet of customer cars to obtain both driving data and road data.
Self-driving vehicles learn how to operate through deep learning systems. The patent refers to the performance as 'limited at least in part by the quality of the training set used to train the model'.
The collection, curation and annotation of the data utilised in creating training sets are tedious and requires significant resources. Moreover, specific use cases that target areas where the machine learning model requires improvement are challenging to collect. Therefore, the system-defined within the patent includes classifying fleet data before it is uploaded to the training model. Essentially, choosing what data will help the system learn.
When will autonomous driving be available commercially?
While some will agree that vehicles with autonomous functionality will be operational within the decade, fully autonomous vehicles that do not require the attention of the driver at all and can operate at a worldwide level are still decades out of reach. However, with the like of Tesla's patent, perhaps it is closer than we think.