Automated driving: The 5 key technical challenges
What’s the first challenge that comes into your head regarding automated driving? Bet you’re not thinking of system architecture or cluster connectivity? Luckily, AD engineers are. And they’re working tirelessly to overcome them.
The challenges most of us think of along the road to fully automated driving are probably things like infrastructure, legal issues and acceptance. It is of course imperative that these are overcome - but for the moment, put on your engineering hat as we address the five main technical challenges that AD is facing.
1. Sensor technology
The challenge: For zero-accidents to become a reality, sensors must be able to paint a highly accurate and reliable picture of a car's surroundings in real time - and in all conditions, including bad weather. On top of that, the technology must be affordable and scalable.
The solution: Various sensor technologies are used in production cars already, for automated features like parking assist or adaptive cruise control.
One such technology is LiDAR (Light Detection And Ranging) which is thought to be a key enabler for fully automated driving thanks to its Hi-Res precision. LiDAR, as its name suggests, is a surveying technology which bounces light beams off objects rather than using radio waves, as with radar. LiDAR's performance in low light or indeed snow has recently been tested by Ford with promising results.
Perhaps the biggest challenge lies in the field of sensor fusion, where input from all of the vehicle's various sensors is brought together and processed. Here, Artificial Intelligence (AI) is a possible solution. Developers are already using deep learning models to help vehicles better understand what's happening around them e.g. recognizing pedestrians and anticipating their movement patterns.
2. Cluster connectivity
The challenge: The internet will have to become the car's sixth sense. To navigate safely on their own, cars will have to be able to practically see around corners. That will only be possible if there is a powerful backend that provides highly accurate traffic information coupled with shared sensor data from other road users.
The solution: What we are essentially referring to is V2X communication. This technology enables vehicles to automatically send out information relating to a wide range of road conditions. Other vehicles or indeed cloud-based, backend systems will be able to receive and process the data. Short-range, direct vehicle-to-vehicle (V2V) communication, for example, is already under development for time-critical functions. Mandatory data packages with road-relevant information are broadcast between vehicles over a licensed Wi-Fi band. This will ensure that if a car suddenly brakes, the vehicles behind will be alerted in a matter of milliseconds.
A powerful backend, however, is also imperative to cluster connectivity for less time-constrained situations. We're talking, of course, about detailed, dynamically updated maps. Indeed, many suppliers and OEMs are heavily investing in HD mapping companies. Take Bosch for example, who recently teamed up with TomTom, or the Audi, BMW, Daimler buy-out of Nokia's HERE maps. Continental have also developed the eHorizon which provides the vehicle with real time, dynamic information from a server - for when sensors are obstructed. As a result, mobile internet technology (e.g. 5G) becomes even more important for ensuring an ever-present network.
3. Human-machine dialog
The challenge: "The evolution of driver assistance to automated driving results in new tasks and challenges for the Human Machine Interface" (Helmut Matschi, Head of Continental's Interior Division). One such challenge is the handover from driver to automated-driving mode and, most importantly, vice-versa. Another being the dialogue between automated vehicles and other road users - be it other manually driven cars, or even pedestrians.
The solution: The Human-Machine Interface quickly and clearly provides the driver with the information he needs to understand and properly react to the actions of the automated vehicle. This may include augmented reality windshield displays as well as voice and gesture controlled solutions. Intuitive Human-Machine Interfaces will be essential for building trust in the new technology.
Many companies are now taking a more holistic system-overarching approach to replace the "dashboard instrument" concept. "A future Holistic Human-Machine Interface will bring joy to everyday driving: Drivers will be able to receive all the information they need. They will be in a constant dialogue with their vehicle - even without words," says Matschi. This is especially important in situations where the vehicle needs to prompt the human driver to reassume driving control. For this, systems using interior cameras which can measure the driver's head position and line of sight are being developed. A line of LEDs for example would redirect the attention to where it is needed.
4. System architecture
The challenge: Automated driving is thought to require one gigabyte of sensor data per minute to be processed in real time. Just imagine; data wise, that's the size of one feature film every second! Future system architectures must integrate all the onboard sensor systems and securely manage this huge amount of information. And as is inevitable, the amount of data required to be processed will only increase in the future.
The solution: Current driver assistance systems often work in an isolated manner: each camera or radar has its own specific function and electronic control unit. However, developers are already starting to speak in terms of a more centralized and flexible architecture which highly automated driving demands. This would migrate many of the AD functions to a central control unit - the brain if you will. It would then combine data from sensors, assess the vehicle's environment and make appropriate control decisions in real time to send to the engine, braking and steering actuators.
This central electronic control unit must however have enough "brainpower" to meet the increased demands. In that space, companies like NVIDIA continue to develop in-car artificial intelligence supercomputers such as the Drive PX2 (presented at the CES 2016) which boasts the processing power of 150 MacBook Pros.
The challenge: When the fully automated driving mode is in operation, the vehicle's occupant is not expected to monitor the system at all. The vehicle must therefore be able to act safely in the event of any malfunction.
The solution: The notion of redundancy is central to the development of fail-safe systems. Suppliers like Bosch are already developing redundant braking systems that could be called upon should the primary system fail. Redundant steering systems that perform targeted braking interventions at individual wheels can safely decelerate the vehicle to a standstill while also keeping it in lane. The challenge for automakers is trying to find the right balance between cost, weight, complexity and robustness for these redundant systems. Reliability does not only apply to driving functions however. Indeed security by design is the basic need here. For example, protection against manipulation attempts must be integrated into the system design from the outset and effective intrusion detection systems must be updated regularly. Hence the importance of the car's software being able to receive real-time updates reliably.
Which do you think is the biggest challenge facing AD technology suppliers? Can you think of any others?