lassedesignen - Fotolia
The launch of Nvidia’s latest chip designed for fully autonomous vehicles shows that the technology behind driverless cars is gathering pace.
The firm’s new Pegasus processor, unveiled last month, offers 10 times more power than its predecessor – more than 320 trillion operations per second – in a significantly smaller package. The design is intended to support so-called “level 5” autonomy in vehicles.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
The Society of Automotive Engineers categorises car automation in six levels, from no automation, or level 0, to full automation, level 5. The highest classification means the system performs steering, acceleration/deceleration, monitoring of the environment and dynamic driving.
Level 5 cars have laser sensing – known as lidar – radar and ultrasonic sensors, which measure an object’s distance away and avoid collisions. These sensors work with on-board cameras to ensure the car has a full awareness of its surroundings. The vehicles also use neural networks to understand and learn what to do in relation to what is around them.
Danny Shapiro, senior director of automotive at Nvidia, told Computer Weekly that as the artificial intelligence (AI) in the car’s neural networks learn, the decision-making process will become better.
“In autonomous vehicles, there is no way to program the decision-making process as there are simply too many possible scenarios that can occur,” he said. “Instead, artificial intelligence is used to handle the uncertainty present when driving.
“The more DNNs [deep neural networks] that are trained through real-world and simulated testing, the safer, the more intuitive and natural a vehicle will behave.”
Shapiro said the process of the car drawing in data from the various sensors, GPS and high-definition mapping is known as “sensor fusion”. Data is processed through Pegasus to decide whether the car should move or not.
Nvidia’s new Drive IX software development kit (SDK) allows developers to build applications for vehicles with any level of automation, opening up new functions to improve passenger safety, said Shapiro.
Read more about driverless cars
- OpenText survey finds that two-thirds of people expect more driverless cars than normal cars in the UK in the next 15 years.
- An explanation of the Fog architecture and why it is crucial for autonomous vehicles.
- BMW and Fiat have partnered with Intel and Mobileye to create an autonomous driving architecture.
“Some of the capabilities include facial recognition, eye tracking, natural language processing and augmented lip-reading,” he said. “These technologies can be used for customisation and personalisation of the driving experience. Cars will automatically unlock, set preferences and start, based on facial recognition.
“The user experience inside the vehicle will be enhanced with natural language understanding and lip-reading. Plus drivers, passengers and pedestrians/cyclists will be safer as the vehicle is able to monitor its surroundings and provide alerts for potential hazards that the driver might not see.”
Shapiro said Nvidia and its partners will have to come up with new technology to keep pace with driverless cars as they continue to develop. Sensor and camera capabilities will advance and the car’s computer will need to improve accordingly to manage all the data.
Not only will there be technical improvements, but the design of the vehicle will also change. Shapiro said future vehicles will resemble “living environments”.
“Without the need to be fully engaged with a conventional vehicle, there is no need for a rigid design,” he said. “In fact, designers are changing the interiors of vehicles to look more like a hotel room on wheels, or an office, or even a high-tech living room.
“Vehicles will have seats that swivel or tuck into the floorboards, ambient lighting will cloak the interior, HUDs [heads-up displays] will be available throughout the vehicle, and in-vehicle entertainment similar to what you expect in a movie theatre or at home will keep passengers entertained.”
As with so many data-rich technologies, driverless car manufacturers will also need to consider consumer concerns, such as privacy. In the future Shapiro sees, the car will track the occupant’s lip movements to understand speech, for example.
“With this system, we bring much of the processing inside the vehicle, instead of having to transmit data to the cloud,” he said. “Also, we utilise proven techniques for cyber security, such as encryption, authentication and virtualisation. In fact, we enable using AI to monitor systems inside the vehicle, so that anomalous behaviour or network traffic can be detected.”
Andrew Lee, head of market intelligence and analysis at Octo Telematics, which provides tracking technology for pay-as-you-drive insurance policies, said data stored in a car should be less of a privacy concern than data on a smartphone, for example.
“Stored data will be essential to allow the system to ‘learn’ and ultimately improve the user experience,” he said. “While we would expect stored data to have cyber security, privacy concerns would sit with the terms and conditions.
“As long as the use of data is for service improvement, you will see little consumer resistance. To put this into perspective, the same terms on a smartphone should be more of a concern for consumers given the continuous use and data stored on it compared to any vehicle.”
With cars becoming increasingly dependent on software, autonomous vehicles will share another challenge with smartphone or desktop technology – how and when to upgrade to the latest versions.
Lee said drivers would need to update the software continuously as the vehicles ages. “On the software side, this is where the improvement will be continuous and it will need to be [transmitted] over the air,” he said. “Updates will be tiered into categories such as ‘critical’ and ‘optional’. Critical updates must be performed frequently and on time, such as insurance and MoT, and optional updates will ensure improvements to the experience and better services.”
Lee predicted that commercial Level 5 vehicles could be on the streets as soon as 2020 – but private cars will come about five years later because of the quicker return on investment for commercial vehicles.