Original equipment manufacturers can overcome technology gaps in the testing processes of autonomous vehicles with the following strategies.
One of the greatest challenges in developing autonomous vehicles (AVs) is ensuring that automotive radar sensors ‘see’ the intended scenarios through software simulations early during the development cycle.
Developing autonomous driving algorithms is complex because the task involves many variables, including replicating complex, repeatable scenes in a lab environment.
The more accurate the scenes, the faster the algorithms can be trained. This is where current in-lab solutions fall short.
The evolution of radar sensors
Over the last decade radar sensors have evolved significantly. In the automotive industry, radar sensors are a critical part of Advanced Driver Assistance Systems (ADAS) and enable features such as blind-spot detection, lane departure warning or correction, automatic emergency braking, and so much more.
Radar technology continues to evolve with higher frequencies, wider bandwidth, and better resolution.
In fact, advances in radar sensor technology push the automation level in vehicles to level 3+ or 4, requiring the test and validation of more and more scenarios.
As a result, automotive original equipment manufacturers and tier-one suppliers need to perform more tests with a higher degree of complexity.
Reimagining test tactics
Picture a setting in an urban area with high population density, many road intersections and turning scenarios.
There are numerous pedestrians, cyclists, e-scooters, and even the three-wheeled cargo delivery bike.
The conventional way of testing the functionalities and algorithms of the radar sensors is by driving on roads for millions of miles. But this tactic will not be able to cover all the potential scenarios, including the one-in-a-million scenarios.
In fact, the majority of the tests that are needed to develop and validate AV systems need to go through simulation much earlier in the development cycle.
Automotive OEMs need to emulate real-world scenarios that enable validation of actual sensors, electronic control unit code, AI logic and more. Testing the physical hardware in a simulated environment close to real-world scenarios ensures that autonomous vehicles will behave as expected on the road.
The simulated environment and the rendered conditions need to include vehicle dynamics, weather conditions, and surrounding objects as well as real-time responses, in order to test the responses of the radar sensors.
However, gaps remain in the technology today that hinder real-world scene renderings.
- Number of targets and field of view
Some systems use multiple radar target simulators (RTS). Each presenting point targets to radar sensors and simulates horizontal and vertical positions by mechanically moving antennas around. Mechanical automation slows overall test time. Other solutions create a wall of antennas with only a few RTS. These solutions only allow the radar sensor to see a handful of objects within a very narrow angle in front of it where blind spots can occur. - Minimum distance
Realistic traffic scenes require objects to be emulated very close to the radar unit. For example, approaching a stoplight where cars are two meters or less apart, bikes or scooters may move in the lane, and pedestrians may cross the road very near to the car. Passing this test is of utmost importance to the ADAS autonomous vehicle’s safety features. - Resolution between objects
The ‘resolution between objects’ metric refers to the details of the scene and the confidence to know that the algorithm tested can distinguish between two objects that are close together. If the objects cannot be identified correctly, it will be difficult to fully test the sensors, the algorithms, and the decisions that rely on the data streaming from the radar sensors.
The robustness of autonomous driving (AD) algorithms depends on how comprehensive the testing is. There is a need to enable original equipment manufacturers in the automotive industry to test autonomous drive systems with radar sensors faster and with highly complex, multi-target scenes.
By thoroughly testing decisions earlier in the cycle against complex, repeatable, high-density scenes, and with stationary objects or objects in motion, automotive manufacturers can accelerate the insights that come from ADAS or AD algorithms.
