The self-driving, autonomous vehicle has been getting lots of attention due to significant development efforts and dramatic progress made by companies such as Google. While the widespread use of autonomous vehicles on public roads is probably still years away, these vehicles are already being employed in ‘constrained’ applications such as open-pit mines and farming.
Among the many technologies that enable autonomous vehicles is a combination of sensors and actuators, sophisticated algorithms, and powerful processors to execute software. The sensors and actuators in an autonomous vehicle fall into three broad categories:
- Navigation and guidance (where you are, where you want to be, how to get there)
- Driving and safety (directing the vehicle, making sure it operates correctly under all circumstances, and follows the rules of the road)
- Performance (managing the car’s basic internal systems).
There are dozens of sub-systems and hundreds of specialised sensor channels under these three categories. Let’s look at a few of them.
Goal No. 1: Know where you are going
Navigation has been a human concern since the earliest days. It addresses two related questions: where you are now, and what paths are available to get to where you want to be. Instruments and techniques such as the compass, sextant and dead reckoning in the past, as well as the more current LORAN radiolocation, are examples of what has been used, with varying degrees of accuracy, consistency and availability.
For the autonomous vehicle, the navigation and guidance sub-system must always be active and checking how the vehicle is doing, in relation to the final goal. For example, if the original ‘optimum’ route has any unexpected diversions, the path must be re-computed in real-time to avoid going in the wrong direction. Since the vehicle is obviously constrained to roadways, this takes much more computational effort than simply drawing a straight line between A and B.
The primary sub-system used for navigation and guidance is based on a GPS (Global Positioning System) receiver, which computes the present position based on a complex analysis of signals received from at least four of the constellations of over 60 low-orbit satellites. A GPS system can provide location accuracy to the order of one metre (the actual number depends on many subtle issues), which is a good start for the vehicle. Note that, for a passenger who plans to hop into the car and get going, a GPS receiver takes between 30 and 60 seconds to establish the car’s initial position, so the autonomous vehicle’s departure is delayed till this first fix is computed.
GPS sub-systems are now available as sophisticated System on Chip (SoC) ICs or multi-chip chipsets which require only power and antenna, and include an embedded, application-specific compute engine to perform the intensive calculations. Although many of these ICs have an internal RF preamp for the 1.5GHz GPS signal, many of the vehicles opt to put the antenna on the roof with a co-located low-noise amplifier (LNA) RF preamplifier, and locate the GPS circuitry in a more convenient location within the vehicle. The antenna must have right-hand circular polarisation characteristics (RHCP) to match the polarisation of the GPS signals, and can be a ceramic-chip unit, a small wound stub design, or be of another configuration.
An example of a GPS module is the RXM-GPS-F4-T from Linx Technologies. This 18mm×13mm×2.2mm surface-mount unit requires a single 1.8V supply at 33mA, and can acquire and track up to 48 satellites simultaneously; more channels allow the GPS to see and capture more data and thus yield better results and fewer dropouts. Its sensitive front-end requires a signal strength of just -159.5dBm for operation. After it computes locations based on the GPS received signals, it provides the output data to the system’s processor via serial interface using the industry-standard National Marine Electronics Association (NMEA) message format.
While GPS is an essential function for autonomous vehicles, it’s not sufficient by itself. The GPS signal is blocked by canyons, tunnels, radio interference, and many other factors, and these outages can last for many minutes and longer. To supplement the GPS, the autonomous vehicle uses inertial guidance, which requires no external signal of any type. The inertial measurement unit (IMU) consists of a platform fixed to the vehicle, which has three gyroscopes and three accelerometers, with one pair of each oriented for the orthogonal X, Y, and Z axes. These sensors provide data on the rotational and linear motion of the platform, which then is used to calculate the motion and position of the vehicle, regardless of speed or any sort of signal obstruction. Note that an IMU cannot tell you where you are, but only tracks the motion, so the initial location of the vehicle must be determined by GPS or entered manually.
The in-vehicle IMU would not be practical without the development of MEMS-based gyros and accelerometers. The historical and fully refined IMU is based on spinning-wheel gyros and a gimbaled platform, which has served many applications quite well (missile guidance/space missions, etc), but it is simply too large, costly, and power-hungry for an autonomous vehicle.
A representative MEMS device is the A3G4250D IC from ST Microelectronics, a low-power 3-axis angular rate sensor which provides a high degree of stability at zero-rate level and with high sensitivity over temperature and time. It provides 16-bit digitised sensor information to the user’s microprocessor via a standard SPI or I2C digital interface, depending on the version chosen. With its tiny size of just 4mm², operation from a 1.8V supply, and stability and accuracy specifications, it is well suited for inertial automotive navigation when combined with a 3-axis accelerometer, for a complete 6-axis IMU.
Goal No. 2: See where you are going
The autonomous car must be able to see and interpret what’s in front when going forward (and what’s behind when in reverse, of course). It is also necessary to see what is on either side; in other words, it needs a 360 degree view. An array of video cameras is the obvious choice, with a camera to determine where the lane is, and to sense objects or markers on the road.
But using cameras alone presents problems. First, there are mechanical issues of setting up multiple cameras correctly and keeping them clean. Second, heavy graphics processing is needed to make sense of images; and third, there is a need for depth perception as well as basic imaging. Finally, conditions of lighting, shadows, and other factors make it very challenging to accurately figure out what the camera is seeing.
Instead, the primary ‘vision’ unit on the autonomous vehicle is a LIDAR system, short for Light Detection and Ranging (or a mash-up of Light and Radar, depending on the source you check). To enable the split-second decision-making needed for self-driving cars, the LIDAR system provides accurate 3D information on the surrounding environment. Using this data, the processor implements object identification, motion vector determination, collision prediction, and avoidance strategies. The LIDAR unit is well-suited to ‘big picture’ imaging, and provides the needed 360 degree view by using a rotating, scanning mirror assembly on the top of the car.
LIDAR provides raw information using high-speed, high-power pulses of laser light that are timed with the response of a detector to calculate the distance to an object from the reflected light. An array of detectors, or a timed camera, can be used to increase the resolution of the 3D information. The pulse is very short to enhance depth resolution, and the resulting light reflections are used to create a 3D point-like ‘cloud’ that is analysed to transform the data into volume identification and vector information. The transformed result is then used to calculate the vehicles’ position, speed, and direction relative to these external objects, to determine the probability of collision, and give instructions for the appropriate action, if needed.
For close-in control, such as when parking, lane-changing, or in bumper-to-bumper traffic, the LIDAR system is not as effective. Therefore, it is supplemented by radars built into the front and rear bumpers as well as the sides of the vehicle. The operating frequency for this radar is usually 77GHz, which has been allocated for this purpose. It has good RF propagation characteristics, and provides sufficient resolution.
To fit the radar into the flat bumper assembly and its limited space, it is necessary to use a highly integrated design, including parts of the radar subsystem PC board as its antenna. Also required are active components such as the AD8283 from Analog Devices, which integrates six channels of a low noise preamplifier (LNA), a programmable gain amplifier (PGA), and an anti-aliasing filter (AAF) plus one direct-to-ADC channel, with a single 12-bit analogue-to-digital converter (ADC).
The primary application for the AD8283 is in high-speed ramp, frequency modulated, continuous wave (HSR-FMCW) radar. The performance of each functional block is optimised to meet the demands of this radar system with a careful balance among parameters such as LNA noise, PGA gain range, AAF cut-off characteristics, and ADC sample rate and resolution. The AD8283 includes a multiplexer in front of the ADC, which automatically switches between each active channel after each ADC sample has been taken. Each channel features a gain range of 16dB to 34dB in 6dB increments and an ADC with a conversion rate of up to 72MSPS. The combined input-referred noise voltage of the entire channel at maximum gain is 3.5nV/√Hz, which is a critical threshold parameter for effective performance.
Goal No. 3: Get to where you are going
While components and sub-systems used for navigation and guidance or for image-capture and sensing get the most attention, a large portion of the design of an autonomous vehicle involves mundane issues such as power management. Several application-specific, unique circuit boards and sub-systems are added to a conventional vehicle to provide the functions needed for autonomous operation. Much of the system-level operations involve measuring and managing the power requirements to control power, overall consumption, and thermal dissipation.
Monitoring the current and voltage at the batteries often requires isolated sensing, for safety and functionally, but isolation is not needed on low-voltage circuit boards. Instead, the most common technique used to determine current at a source or load is with a high-side, current-sense, milliohm resistor (called a shunt), in conjunction with a differential amplifier that measures the voltage drop across it. Although the amplifier is used with a discrete sense resistor, there is now an alternative that saves space, minimises errors in readings which are primarily due to thermal drift of the sense resistor as it self-heats, and simplifies the bill of materials (BOM) by reducing the number of parts. The INA250 from Texas Instruments puts a sense resistor and differential amplifier in a single package, resulting in a far smaller board-layout footprint, fewer circuit-layout problems, and lower system costs due to a simplified schematic.
The autonomous car has attracted a great deal of interest (and scepticism) as well as considerable R&D investments. How practical or affordable it will actually be, or when we’ll see it as a mainstream vehicle, is unknown and the subject gives rise to much speculation. There has been significant progress, demonstrated by millions of test miles on public roads, to refine its design and operation. We do know that such a vehicle demands a complex integration of sophisticated algorithms running on powerful processors, making critical decisions based on large streams of real-time data coming from a diverse and complex array of sensors.
Bill Schweber is an electronics engineer —MSEE (University of Massachusetts) and BSEE (Columbia University)—and has written three textbooks on electronic communications systems, as well as hundreds of technical articles, opinion columns, and product features. He has also planned and presented online courses on a variety of engineering topics. He has worked as a technical website manager for multiple topic-specific sites for EE Times, as both the executive editor and analogue editor at EDN, as well as in both the client and publication sides of public relations (PR).