Dead Reckoning is not Mission Impossible: Functional Safety Challenges in ADAS and Autonomous Vehicles

Over the last few years, Lorit has been very active with sensors used in autonomous vehicles (AVs) and advanced driver assistance systems (ADAS) of many shapes and sizes – from trucks and cars to robots and tractors. There are many challenges associated with sensors, not only deciding which ones to utilize in the application, but also how to detect and mitigate faults when implementing the chosen sensor technology.

There are Safety Integrity Level (SIL) requirements for sensors in general, and in the automotive domain this translates to the Automotive Safety Integrity Level (ASIL) defined by ISO 26262. The key challenge being whether a sensor can realize the required safety integrity on its own, or if additional redundant technologies are required to meet these functional safety demands.

In this blog we look at some of the issues and solutions to mitigate sensor faults in autonomous and ADAS applications.

GNSS Signal Loss in Tunnels and the Role of Dead Reckoning

Many vehicles depend on positional information to pinpoint exactly where they are at any given moment in time. Just as we are all accustomed to poor Wi-Fi reception when working in cafés, a weak or lost signal can have far more significant impact if a vehicle relies on this positional data while moving at speed.

Global Navigation Satellite Systems (GNSS) are used widely nowadays to provide autonomous geo-spatial positioning. This satellite data along with other diverse sensor inputs can achieve impressive accuracy – ranging from one meter down to centimeter-level precision, depending on the technology.

But what happens when a vehicle enters a tunnel, some many kilometres long, and the GNSS signal is attenuated to the extent the signal is no longer reliable?

This is where dead reckoning comes into play. By combining speed and directional information, the vehicle’s location can be estimated during periods of reduced GNSS quality. Automotive Dead Reckoning (ADR) integrates data from a 6-axis Inertial Measurement Unit (IMU) to estimate position changes based on speed and direction.

Dead reckoning, utilizing 3-axis accelerometers and 3-axis gyroscopes to estimate the position of a moving object, is a common technical solution nowadays. Starting from a known reference point, it calculates changes in position by factoring in speed, travel direction, and elapsed time – allowing vehicles to maintain reliable positioning even at the point when GNSS signal is unavailable.

Blinding Sensors: Object Detection and Functional Safety in ADAS

Many different sensor technologies are used for object detection, both internal and external to a vehicle. At Lorit, we have worked on a number of Time-of-Flight (ToF) sensors, which can detect who is sitting within a vehicle. This eliminates the reliance on pressure sensors to determine seat occupancy. Moreover, such ToF sensors, as 3D imaging sensors, can identify exactly where the driver/passenger is positioned, enabling optimized control of airbag inflation pressure. Because of this, ToF sensors often carry a fairly high safety integrity requirement.

ToF sensors can, however, suffer from defective pixels that could result in obscuring objects. “Dummy” pixels can be used as test pixels to confirm the proper operation of the sensor. The size of an object in relation to the number of defective pixels also affects what might be potentially missed.

Additional issues can arise due to the analog nature of these sensors: unambiguous ranges, signal-to-noise challenges, saturation, and “wiggling” (waveform deviations in the sensor). These also have optimum correction criteria.