CC Blog Insights Tech The Future

The Future of Autonomous Cars

Working with a suite of sensors, IMU Sensors can help ensure safe operation for autonomous cars.
Written by James Fennelly

Sensors, Software and More Sensors

Everyone keeps trying to predict when self-driving cars will be here—2020, 2025, 2030 and so on. In fact, Level 4 vehicles are here already [1]. Just go to Phoenix Arizona and watch a Waymo One subscriber hail a self-driving ride share vehicle. There are dozens of them operating on the roads throughout the metro Phoenix area. The challenge now is to achieve Level 4 at a cost that is not prohibitive and reaches the quality, reliability and longevity levels needed for the high-volume mass-production automotive market.

Sensors and software are the key technologies, but as of yet, the particular sensor suite solution needed for Level 4 and Level 5 implementation is still being debated. Most notably Elon Musk and Tesla lead the camp that believes LiDAR (light detection and ranging) sensors are NOT needed, while seemingly all others think LiDAR is essential.

Of the various sensor options, LiDAR sensors are the most expensive. Mr. Musk’s contention is that humans do not need LiDAR to drive, so it should be possible to self-drive without LiDAR. In fact, at Tesla’s recent Autonomy Investor Day, Mr. Musk indicated that Tesla’s Full Self-Driving system (without LiDAR) will be feature complete by the end of 2019 and it will be robust enough to permit users to not pay attention while using it by Q2 of 2020. Table 1 lists the majority of the sensors being used in self-driving vehicles with a short description of their primary functions, strengths and weaknesses.

Table 1: Sensors for self-driving vehicles
Table 1: Sensors for self-driving vehicles

The Role of GNSS

Of note, GNSS (Global Navigation Satellite System – GPS, GLONASS, BeiDou, Galileo and so on) will certainly play a part. There are a couple of enhancements or extensions to GNSS called RTK (Real Time Kinematic) and PPP (Precise Point Positioning) which greatly improves the accuracy of GNSS from a few meters down to a few centimeters. It is unclear if standard GNSS will be used to get close to real position and then the sensors on the vehicle will be used for further refinement, or if either RTK or PPP will be needed to provide the precise positioning. RTK and PPP solutions are much more costly but new chipsets are being introduced which show promise of driving the cost down to the levels needed for the mass market.

All of these sensor technologies have some blind spots. The overlap of the sensor’s capabilities and fusion of the data is critical to system functionality. For example, if the LiDAR is temporarily blinded by atmospheric conditions, the radar, IR (infrared) and visual light camera information can be used to maintain safe operation until conditions for the LiDAR improve. However, conditions certainly can occur that will result in insufficient data from the available sensors for safe operation. Furthermore, there is still the question as to how well the LiDAR will work when there are many LiDARs blasting light at oncoming cars also equipped with LiDAR. This is analogous to traffic coming at you with their high-beam lights on. Your view is greatly reduced.

Often overlooked but critical to supplementing these sensors are IMUs (Inertial Measurement Units). An IMU measures 3 dimensions of linear acceleration and 3 dimensions of rotational rate. From this information, attitude (pitch and roll), change in heading, velocity and position can be calculated. IMUs are used to supplement or fill in the gaps between GNSS updates and can even dead reckon position during more prolonged outages of GNSS and other sensors in the suite.

IMUs are also used to check that the vehicle movement is consistent with the driverless system input (for example, steering wheel position, brake position, accelerator position and wheel speed sensors) to detect if the car is sliding, skidding or in some other out-of-control condition not consistent with inputs, so action can be taken to bring the vehicle back under control (Figure 1).

Figure 1: Working with a suite of sensors, IMU Sensors can help ensure safe operation for autonomous cars.
Figure 1: Working with a suite of sensors, IMU Sensors can help ensure safe operation for autonomous cars.

IMU Advantages

The key strength of the IMU is that it works the same in all weather and geographic conditions. It is self-contained and does not suffer from degradation or outages due to weather, mud on lenses, multipath of radar and LiDAR returns or Urban canyon effects. It is an independent data source that can be used for short term navigation and to corroborate information from other sensors. Because the sensor is independent and immune to all the other interference errors, the current trend is to view the IMU as not only a sensor to supplement and corroborate other sensors, but also to be the sensor of last resort, used to safely maneuver the vehicle out of traffic and bring the vehicle to a stop in a controlled manner when too many of the other sensors have failed or are impaired.

All vehicles in the market today equipped with ESC (electronic stability control) already have a low cost IMU (less than $10) integrated within. However, low cost IMUs are not accurate enough to be of much benefit for inertial navigation. Precision IMUs can easily meet the performance requirements needed for inertial navigation, but today, they can cost many thousands of dollars making them not feasible for wide scale deployment in the Automotive Market. Fortunately, as is with the case of LiDAR, many companies are working on bringing the cost of IMUs with performance adequate for short term navigation into the under $25 range. The required level of raw IMU performance is still not completely determined.

This uncertainty arises because many of the error sources for IMUs can be reduced using data from the other sensors and the target requirement is expressed not in terms of IMU performance but instead in terms of position accuracy versus time derived from the IMU data. Some commonly expressed targets are 30 cm after 30 seconds traveling at under 30  km/ hour and 20 cm after 90 seconds. There are just so many variables associated with…

  • The algorithm and sensors being used to correct IMU errors.
  • The algorithm calculating the position and velocity from the IMU data.
  • The initial condition of the vehicle when the dead reckoning starts (velocity, turning or going straight, slowing or speeding up, going uphill or downhill and so forth).
  • The path the vehicle takes after the dead reckoning starts.

…that it is difficult for the OEMs to translate this into IMU specifications.

Undoubtedly self-driving passenger cars will be equipped with many, many sensors. IMUs without question will be included. We can expect some refinement of the IMU specifications soon. In the next 9 to 18 months we can also expect to learn if Mr. Musk is successful in making a Level 4 solution available without LiDAR, and if LiDAR and GNSS (RTK or PPP) are cost reduced enough to be viable in the consumer automotive market. There is a lot of energy and money being directed in these areas so I expect significant progress to come. Fortunately, we will not have to wait very long to have answers. It is an exciting time in the auto industry. Just don’t blink, because you might miss something important.

RESOURCES

Published in Circuit Cellar Magazine Issue 348 • July 2019 — Get a PDF of the Issue.


Don't miss out on upcoming issues of Circuit Cellar. Subscribe today!

 
 
Note: We’ve made the October 2017 issue of Circuit Cellar available as a free sample issue. In it, you’ll find a rich variety of the kinds of articles and information that exemplify a typical issue of the current magazine.


Would you like to write for Circuit Cellar? We are always accepting articles/posts from the technical community. Get in touch with us and let's discuss your ideas.

Sponsored By

James received his BS EET from the University of Massachusetts. He has been working for the past 10 years with MEMS inertial sensors including component level acceleration sensors and system level products. He is responsible for defining new products at ACEINNA Inc. to meet the needs of emerging applications in the inertial sensing market.