The focus of Continental’s presentation to a press gathering at its test facility in Brimley, MI, last week was a series of new ADAS (advanced driver assistance system) features, including both preproduction systems and more advanced prototype concepts. Some were incremental advancements of existing technology. These included an advanced Trailer Merge Assist and Autonomous Emergency Braking (AEB) for light-duty trucks. Trailer Merge Assist is an advanced blind-spot detection that includes a towed trailer, especially useful for drivers who might tow a trailer only a few times a year.

Other technologies on display also highlighted some of the core difficulties developers are discovering with ADAS. The existing set of 24-GHz radar, lidar, and sonic sensors seems to provide enough data to sense the environment and detect potential dangers. Presenting that data as information for the driver to act upon, through the human-machine interface (HMI), is a key challenge. Most systems to date sense an impending crash, process the data, and actuate preventive actions: stopping, slowing, or steering the car to avoid catastrophe. (They could sense, plan, and act entirely with future autonomous systems.)

The next step is to sense the answer to a crucial question: is the driver even paying attention? If so, does the system need to react?

Driver attentive models

Why is knowing if the driver is attentive so important? Overly restrictive interventions can frustrate aggressive drivers. Jeremy McClain, Engineering Manager-Advanced Technology in the company’s Chassis & Safety Division, gave an example of a driver of a car equipped with AEB. If the driver is quickly overtaking a slow moving car he/she intends to pass, the AEB may prematurely activate a collision avoidance ADAS or sound off unneeded warnings. Such false alarms make drivers ignore or disable their ADAS after too many such irritations. If the system could predict the intentions of the driver, it just may lead to fewer false alarms and a more effective ADAS. “We need to bring the driver more into the loop in an active safety system,” explained McClain.

To help build that solution, Continental showed off a Driver Focus Vehicle demonstrator equipped with an infrared camera mounted in the steering column pointed at the driver. “We make an estimate of driver attention based on where the driver’s face is pointed,” he said. The concept is to adjust a forward collision AEB based on driver attention. When the system senses that the driver is looking away while rapidly approaching a slower moving car, it estimates that the driver is not paying attention. In that case, brakes are applied sooner and a warning light attempts to focus the attention of the driver on the impending danger. However, if the system estimates that the driver is indeed paying attention, activation of braking and warnings are delayed until the last moment before a crash occurs, allowing the driver more control. The system was first reported by SAE in early 2013 (see http://articles.sae.org/11842/).

“The driver attention model is a key component of the future systems we are going to develop,” explained Ibro Muharemovic of the Competence Center Automated Driving & ContiGuard-Advanced Technology in an interview. “It is very challenging because it is not binary, it is not a question of ‘attentive’ or ‘not attentive’,” he remarked. A complex model is developed in Mathworks' MATLAB Simulink and then auto-coded for use in driveaway systems. “Of course, that is one of the many tools we use,” he said.

Modeling how humans drive

Another interesting system along the same lines was an ADAS tool for auto-assisted backup. It uses a 185-degree fish-eye camera driven by a probabilistic model of scenario risk. “There are 18,000 injuries and 300 deaths every year that are attributed to drivers backing into pedestrians in the United States alone,” explained Graham Fletcher, Software Engineer. A static backup camera, providing just a picture to the driver, would help with such a grim statistic. However, the estimated effectiveness of it preventing incidents is only 45% according to NHTSA. Why so low? Drivers may not pay attention to a passive display. They may be distracted for other reasons.

Enter an active ADAS solution that would brake the car automatically. To mimic a human driver’s perception of risk and their response to events, Continental uses a map of probable risk gathered from the area behind and around the back of the car. Stationary objects directly in the path generate high probability of risk. Objects near the car, that might be hit if they move into the path, generate a lower probability of risk—allowing the algorithm designers to program the car to slow without stopping. The slower motion either alerts the driver or provides the system more reaction time in case the object starts to move into the path.

Benefits include a smooth, gradual speed change instead of abrupt panic stops. “Our main point is that we are trying to provide a human like approach,” remarked Fletcher. “A human will slow down a vehicle when they become aware of a danger, and protect for that danger. If it goes away, they will speed up. Our ultimate goal is deliver a system that operates so smoothly a passenger cannot tell if it was our [ADAS] system or the driver who intervened.”