Self-driving Cars Have Challenges Ahead
(It’s the future, but we still have some things to figure out)
Since the dawn of the automobile, the nature of transportation has continuously changed and evolved. Despite all of the changes we’ve seen over the decades, one thing has always remained the same: cars need a driver. If current predictions are to be believed, in 10 years that won’t be the case.
Even conservative projections are forecasting 10 million self-driving cars on the road by 2020. Join us as we look at the hardware and software behind self-driving cars, and how new solutions are poised to solve many of the current problems with the technology.
Let’s Start With Sensors
The idea of relinquishing control over our cars is a frightening concept for many people. Car accidents happen every day as a result of human error, so wouldn’t it be safer if we had advanced technology behind the wheel?
In 2016, a Tesla Model S was involved in a fatal accident while the Autopilot was engaged. The cause of the accident was revealed in a Tesla blog post.
In short, the issue was a result of the system not recognizing the white side of a tractor-trailer against a brightly lit sky. As a result, the brakes did not engage. Scenarios like this one reveal the importance of sensors that can distinguish between fine details.
This is where LiDAR is poised to solve a lot of problems. The fundamental concept of this technology is similar to radar, except it uses infrared light instead of radio waves.
By sending out pulses of light and measuring the time it takes to bounce back, this technology can build detailed 3D point-like “cloud” which is then used to identify volume and vector information.
All of this data is used to calculate the vehicles, position, speed, and direction relative to other objects. A typical LiDAR system utilizes four components:
- Lasers: Usually with a wavelength of 1550nm to make them safe for human eyes.
- Scanners and optics: Scanning tools are used to interpret the data and build the point map. These dictate the resolution and range of the system.
- Photo detector and receivers: This device reads and records the signal as it returns to the system. There are two main types: solid state detectors and photomultipliers.
- Navigation and positioning systems: When LiDAR is mounted on a vehicle, it needs the absolute position orientation of the sensor in order to provide useful data. Global Positioning Systems (GPS) can provide accurate information and an Inertial Measurement Unit (IMU).
The LiDAR system is not as accurate in scenarios like bumper-to-bumper traffic. Radars are used in the front and rear bumpers, along with the sides of the vehicle for these situations.
To get all the technology a radar requires into the small space, there is a highly integrated design. Part of this arrangement involves using the subsystem PC board as the antenna.
Active receiver processing components (like the AD8283 from Analog Devices) optimize the radar system. The receiver incorporates a multiplexer in front of the ADC to automatically switch between each active channel when an ADC samples has been taken.
There’s a careful balance when it comes to sensors like radar and LiDAR. Despite the technology being widely available for some time, it’s still too underdeveloped for use on the road.
It is still quite expensive- the cheapest LiDAR sensor on the marketing is the Velodyne puck which costs $4,000. The technology also needs to find a way to stand the test of time across a wide range of driving conditions.
LiDAR in conjunction with radar for close quarters driving seems to be the best approach. The concept is solid, but the technology needs to catch up before it can be widely implemented.
Power Management & Autonomous Operation
While much of the mainstream focus tends to be on navigational sensors and technology, a major issue arises when we look at power management. When developing a self-driving car, new circuit boards and subsystems are added to realize the goal of autonomous operation.
The ability to measure and manage power requirements for consumption and thermal dissipation is paramount. Isolated sensing is often used to monitor the current and voltage of batteries, but it is not a requirement for low-voltage circuit boards.
The most common way to determine current is to use a high-side, current-sense, milliohm resistor known as a shunt in combination with a differential amplifier. The INA250 from Texas Instruments combines a sense resistor and a differential amplifier into a single product. This saves space by being smaller, minimizes errors, and simplifies the bill of materials (BOM) when ordering parts.
Solutions like this one will help us solve problems with power management in a way that optimizes space while also reducing the overall complexity of the solution.
The Software Side of Things
Once we have solved the current issues with sensors, power management, and navigation, we will need a standardized software to make it all work. Machine learning algorithms are used to interpret the data from external sensors and act accordingly.
They are classified as supervised and unsupervised algorithms, the difference being how they learn. Supervised ones use a training dataset to learn until they reach a minimal probability of error.
Unsupervised algorithms seek to gain value from available data. It detects patterns or divides the data into subgroups. For self-driving cars, these algorithms are used to continuously render a surrounding environment and forecast potential changes.
This includes detecting objects, identifying them, and predicting their movement. This type of software could vary between brands, resulting in many approaches to the same problem.
For safety reasons, self-driving vehicles should be able to communicate with each other to reduce the chances of collision. Ideally, the government would require companies to make their self-driving software and protocols public so the various brands can communicate. It’s not a perfect solution, though, as there is no standard yet. Even so, a rule like this would prevent a monopolization of the software used in self-driving cars.
Over To You
Self-driving cars are expected to become the new norm within the next ten years. The primary issues reside on the hardware side of things, but a solution for software is also needed before this technology can truly proliferate.
Are you an engineer or designer working on self-driving car technology? What kinds of problems are you working on? Let us know in the comments!