Autonomous Driving Technology

If autonomous vehicles are to replace human drivers – they need to be able to manage the entire dynamic driving task (DDT).

This article is an introduction to the technology that drives autonomous vehicles. For more general article about driverless vehicles, see the Autonomous Driving Info, Projects & Resources page.

CONTENTS

• Autonomous driving architecture
• Environment Mapping/Localisation
• Perception
• Cameras
• Motion planning
• Control
• Connectivity and communication
• What is the role of Artificial Intelligence (AI) in autonomous vehicles?
• Autonomous vehicle resources
• Autonomous driving technology: Facts and figures
• iMOVE’s autonomous driving works and impacts
• iMOVE autonomous driving projects
• iMOVE autonomous driving PhD projects
• iMOVE autonomous driving articles

Image source: Autonomous Driving Architectures: Insights of Machine Learning and Deep Learning Algorithms

Autonomous driving architecture

The DDT includes all the functions a driver or system needs to safely:

• operate the vehicle including steering, braking and acceleration
• monitor conditions inside and outside the vehicle
• detect and respond to objects and events

This requires a complex technology stack, for example:

1. Sensors that allow the vehicle to perceive its environment – similar to a person’s five senses.
2. Applications to process data from sensors and support decision-making – like the human brain.
3. An operating system that allows someone to interface with the vehicle’s software and hardware components – for example, Windows or MacOSX on your personal computer.
4. Networks the vehicle can access to send and receive information and computation to help the vehicle process it – like how a person would access the internet.
5. Infrastructure to support autonomous driving – both inside and outside the vehicle. This includes systems that control the vehicle’s movement, as well as physical and digital road infrastructure.

Self-driving cars in the future may use a combination of the technology already available today. Or newer and safer alternatives may emerge. Using established technologies can be beneficial because they have been through more testing. Newer solutions will take longer to be approved for commercial use – but may outperform competitors in the long run.

Environment Mapping/Localisation

Automated vehicles use variety of means to localise itself in the environment.

Global positioning system (GPS): Connected and autonomous vehicles have high-precision GPS receivers that help them localise within the environment.

Simultaneous Localisation and Mapping (SLAM): High accuracy localisation can be achieved applying SLAM technique by using data from LiDAR and/or Camera.

Inertial Measurement (IM): Connected and Autonomous vehicles also use Inertial Measurement Unit (IMU) along with GPS and SLAM to support accurate localisation.

Perception

Automated vehicles use a variety of sensors to map and navigate their environment. Sensors allow the vehicle to detect other vehicles, pedestrians, obstacles, and traffic. Better sensors mean better information for the vehicle’s systems to interpret – and improved road safety as a result.

Light detection and ranging (LiDAR): LiDAR uses laser beams to create a detailed 3D map of the vehicle’s environment. If a system is using LiDAR, a spinning sensor that provides 360° data is typically present on the top of the vehicle.

Radio detection and ranging (Radar): Automated driving systems may use radio waves and sensors to detect objects and obstacles. Radar helps to detect the speed and orientation of moving objects and people.

Sound navigation and ranging (Sonar): An automated vehicle might use sound waves to detect objects nearby. Sound waves are most useful for low-speed manoeuvres and parking assistance.

LiDar in action

This short video is of ZOE1, a non-autonomous vehicle used to investigate the infrastructure needs of automated vehicles now and in the future. On-board ZOE1 was 3 x forward‐facing cameras, 1 x 360-degree camera, 1 x roof-mounted 32-layer LIDAR, GPS (Global Positioning System) sensors, and 2 x on‐board data-logging computers.

See a short snippet of simulataneous vision of what the camera sees, and what LiDar sees, at the 13 second mark of the video.

Read more about ZOE1, and the findings of the project it was used for, at Autonomous vehicles and Australian roads: Are they ready for each other?

Cameras

Many modern, non-autonomous cars already use cameras to assist with parking and blind spot detection. Vehicles with higher levels of automation may have onboard systems that interpret and act on camera data. They may use a single camera or stereovision or surround vision – which involves several cameras creating a 360 degree view.

Motion planning

Once the vehicle knows where it is in the environment (via Environment mapping/Localisation) and what is happening around it (via Perception system), its software and applications now use this information along with prior maps of the environment (also known as High Definition Maps) to plan its motion.

Planning algorithms enable advanced autonomous vehicles to plan and make decisions about:

• where to move next and how fast
• how conditions may influence their short-term movement and longer-term route planning
• how to interact with other vehicles and safely respond to their behaviour
• avoiding traffic and collisions based on their own history and experiences with human drivers
• how to collaborate with other autonomous vehicles to move passengers and freight using the most efficient routes and methods.

The main purpose of planning functions is to identify the safest and most efficient way to navigate dynamic driving conditions. Based upon motion planning calculations the vehicle will now interact with onboard systems and hardware, infrastructure and networks to determine the safest way to drive.

Control

Based upon the motion planning calculations the control system decides how much steering angle, brake force and/or acceleration is to be applied to various hardware to achieve outcomes calculated by the Motion planning algorithms.

Connectivity and communication

Advanced automation will rely on wireless communication systems that interact with other vehicles, infrastructures and the environment. Wireless communication may also tap into online networks and devices carried by vulnerable road users.

Currently, researchers are exploring a range of communication possibilities to support the safety of driverless cars:

• Vehicle-to-vehicle (V2V) communication allows vehicles to share data with each other in real time.
• Vehicle-to-infrastructure (V2I) communication interacts with roadside infrastructure like traffic lights and road signs.
• Vehicle-to-cloud (V2C) or vehicle-to-network (V2N) communication allows vehicles to access central servers and networked data, for example maps or real-time traffic data.

There are a variety of technologies that could enable vehicles to communicate this way. A combination of these methods (or new ones) may be used in the future to achieve safe, sustained, and suitable network connectivity.

What is the role of Artificial Intelligence (AI) in autonomous vehicles?

As described above in the autonomous vehicle system architecture section, these vehicles have several systems. More or less all of these systems make use of AI, however Localisation, Perception and Motion planning stages use AI more extensively through their respective algorithms.

• Location algorithms: that estimate where the vehicle is based on GPS and other inputs
• Perception algorithms: that process sensor data and detect as well as classify objects, such as pedestrian, cyclist, car, truck and so on.
• Prediction algorithms: that anticipate behaviour of other road users and the trajectory of objects and obstacles. These predictions are used in motion planning.

iMOVE, Department of Transport and Main Roads (Queensland), and QUT put an autonomous car on Queensland streets

ZOE2, is an Renault Zoe electric vehicle, modified with technology that takes the car up to a SAE Level 4 automated vehicle capability. It has trialled on the Mount Cotton test track, and also on the streets of Ipswich, Bundaberg, and Mt Isa, with members of the public taking the opportunity to take a ride in the car.

Read more about the project, its findings and recommendations, at Cooperative and Highly Automated Driving Safety Study.

We can expect Level 3/Level 4 consumer vehicles with limited Operational design domain, such as on highways, entering the market around 2026, albeit in a very limited number. We could potentially see robotaxi service provider entering Australian market, initially on trial basis.

There are also possibilities of connecting a container freight hub with a port through automated container platforms during off peak hours, thus expanding port’s capacity and utilising excess road capacity during off peak.

Amit Trivedi – Program Manager at Queensland’s Department of Transport and Main Roads’ (TMR) Cooperative and Highly Automated Driving pilot (CHAD)

Autonomous Driving resources

Here’s a selection of Australian strategy and project documents on the topic of autonomous driving.

Autonomous driving technology: Facts and figures

High Definition Maps

• A High Definition (HD) map contains a rich, 3D representation of the world with high positional accuracy. HD maps include all of the features of enhanced digital maps, but with much greater
  accuracy and contains detailed information necessary for automated driving.
• A HD map will typically include lane positions and widths, positions and descriptions of road signs, logic data such as the direction of travel of a lane, and environmental data for the purpose of localisation.
• Creating a HD map is a complex task; it is not just a matter of simply recording sensor data, but also involves adding semantics (meaning) to the data. For example, a traffic sign observed
  with a camera needs to be identified and marked on the HD map.

LiDAR

• LiDAR is particularly powerful when used to generate a 3D representation of the world around a vehicle. These are typically called LiDAR point clouds. LiDAR point clouds are one of the standard methods of representing an environment in high definition across automated vehicles and robots, but have high data storage requirements.
• LiDAR can be simultaneously synchronised with image data. The colours denote different distances to parts of the image. Combining both LiDAR and vision enables improved classification of objects in the environment by combining geometrical and appearance information.

Source: HD mapping Australia’s CAV future: Final report

What is iMOVE doing in the area of autonomous driving?

iMOVE is carrying out R&D in a number facets around driverless vehicles. In a more top-down approach, in What C-ITS technologies for national deployment in Australia and Accelerating the uptake of C-ITS technologies in Australia we’re laying the ground for the introduction of autonomous vehicles in Australia by ensuring we select best technologies for our roads.

That’s a smart, important move, as is the mission of bringing the public along on this transport shift. Promoting community readiness and uptake of CAVs and Cooperative and Highly Automated Driving Safety Study. How do we educate the public on this mode, and how can governments increase community acceptance and confidence? We’re finding put!

Deeper dives into the technologies have been undertaken in HD mapping Australia’s CAV future, Development and use of cooperative perception for CAVs, and Improved sensing for signalised intersections.

Connected and automated vehicles are also important additions for vulnerable road users, and in this area we’re investigating issues and opportunities via 5G aid in automated mobility for elderly and people with disability, and Australia’s Public Transport Disability Standards and CAVs.

What impact iMOVE is having in the area of autonomous driving?

iMOVE and its partners are at the forefront of work in preparing for connectivity and automation of vehicles to help make Australian roads safer.

Our research and development is being used to understand automated driving in an Australian context, including readiness of our road assets (signs, lines etc) are for AVs, how drivers behave in response to AVs, how connected vehicle technologies can be integrated in automated technologies and the general performance of the technology.

While the work we’re doing is taking place in separate locations, what we are learning is highly applicable right across Australia.

Additionally, we’re readying Australia’s next generations of experts and practitioners to help make Australian roads prepare for driverless vehicles via our Undergraduate Student Industry and Industry PhD programs.

Contact iMOVE

There’s still a lot of work to be done to make Australian transport systems safer. If you’d like to talk to us about any R&D work in the area of autonomous driving please get in touch with us to start a discussion.

iMOVE autonomous driving projects

iMOVE, along with its partners, is active in carrying out R&D to advance autonomous driving technologies in Australia.

Please find below the three latest autonomous driving projects. Or click to view all iMOVE’s autonomous driving projects.

iMOVE autonomous driving PhD projects

In addition to iMOVE and its partners’ autonomous driving projects listed above, as part of our Industry PhD Program businesses, universities and PhD students work on an agreed topic over a three-year period.

These are the three most recent PhD projects that have been undertaken on the topic of autonomous driving. Click to view all iMOVE’s autonomous driving PhD projects.

iMOVE autonomous driving articles

In addition to projects, iMOVE also publishes articles, thoughtpieces, case studies, etc. that cover the many issues and solutions around autonomous driving.

Below are the three most recent articles. Or click to view all iMOVE’s autonomous driving articles.