Autonomous Driving

Event Detection is a core function of autonomous vehicles

With advancements in artificial intelligence and related technologies, cars will eventually become autonomous vehicles that are entrusted with human lives. In order for these autonomous vehicles to function properly, they must be equipped with capabilities that enable them to make the best and safest decisions in any situation. To ensure this, the images produced by the vehicles' cameras are analyzed to identify the position of objects, the associated category, and related events to better understand the images and the situations. The systems currently available were inspired by computer vision and deep learning methods. However, they have difficulties when trying to detect small objects, objects with random geometric transformations, and when there is a lack of contrast For this reason, EDI GmbH is working on an algorithm that will improve the detection of critical events by recognizing contexts and correlations in situations where the sensors have weak points. EDI GmbH's algorithm will also be helpful in avoiding accidents in road traffic. The detection of object and events covers a variety of important techniques, especially image processing, pattern recognition, artificial intelligence, and machine learning.

Failure strategies for robot taxis

The AnRox research project aims to develop an optimized drive system for automated electric vehicles. As a project partner, EDI GmbH has the task of developing and validating a substitute system that steps in if the primary system fails Many well-known companies and institutions such as Bosch, Siemens, Infineon, and RWTH Aachen are working closely as partners to develop the efficient and fail-safe electrical system for robot taxis. Using its dynamic risk management algorithm and predictive behavior, EDI GmbH's task is to develop strategies based on existing data to tell the system how to react, if the primary system or parts of it fail In case of a breakdown, the intelligent ADAS (Advanced Driver Assistance System) will assess the situation and will stop the vehicle safely, so that the passengers are not endangered and safe. Moreover, in the event of a breakdown occurring while driving as well, if the vehicle detects an object such as a person or another vehicle in a place where the driver may not be able to notice them, the system will alert the driver. Likewise, if the car's battery becomes weak or the brake does not work properly the intelligent ADAS can alert the driver. If the system determines that the vehicle is leaving its lane, it can activate the lane departure warning. In a nutshell, artificial intelligence is used to make the right as to how the vehicle should react to the error situation so as not to endanger the passengers and to bring the vehicle to a safe stop. The validation of these scenarios is carried out in a simulation.

AI-based Safe Navigation and Driving

Risk Estimation with a Learning AI - RELAI

Dynamic Risk Management (DRM)

Today's functions of autonomous driving systems are limited to assisting the driver or controlling the vehicle in simple, clearly defined situations, such as parking or driving on the highway. Responsibility still lies in the hands of the driver. The autonomous driving systems have not exceeded automation level 3 in series production. EDI GmbH has developed an intelligent algorithm (EDI Dynamic Risc Management) that enables autonomous driving systems to manage various risks on the road as dynamically as experienced and responsible human drivers would.

Dynamic Risk Management (DRM) is an application of artificial intelligence (AI) that is able to constantly observe data streams from different sources, merge them and make a decision depending on the current situation. In the case of automated driving, AI can determine driving behavior that is perceived as appropriate for the driver, the passengers and other road users, in addition to purely evaluating the driving context in terms of safety.
The original application area for our DRM algorithm is autonomous driving. For the development of the DRM algorithm, over 100,000 critical road traffic incidents were evaluated from recorded data with imagery using machine learning. Each incident was then specified manually by traffic experts with over 100 different parameters and in some cases with up to 10 characteristics. Using the trained AI, further recorded journeys can now be automatically evaluated. The relevant parameters are extracted automatically.

 Some of the parameters that have been weighted into different risk levels with the AI are driver behavior, speed history in different situations, existing infrastructure and intersection and road types. Another important aspect considered in the model of our DRM is the behavior of other road users: are they pedestrians, cyclists or other cars? The age of the pedestrians and whether they are drunk or not also plays a role.

Overall, our DRM covers a very large parameter space and the algorithm is correspondingly powerful: it can predict critical situations. Thus, "safety", the 3rd dimension of navigation, is implemented predictably. The algorithm is so robust that an assessment of the situation can be made even if not all parameters are available. The more information there is, the more accurate the statement is, of course.

The required data comes from different sources, such as the digital map of the navigation system and the camera system of the autonomous vehicle, which might perceive many cyclists who are in front of the vehicle for example. In addition, there is data from other sensors of the autonomous vehicle such as radar, ultrasound and lidar and further data from sensors that may be located in the public infrastructure and that can communicate with the autonomous vehicle. 

The approach is transferable, which is why the areas of application for Dynamic Risk Management are not just self-driving vehicles. It is also used in our well-being barometer, which accompanies seniors in their daily lives and notifies relatives and caregivers when there are deviations in the daily routine. A large butcher's shop could also use the DRM to predict when demand for grilled sausages will be particularly high. Weather data and the occurrence of major events play a role here. Thanks to our DRM, the butcher no longer has to rely solely on his gut feeling. He then also has reliable support from the AI. This example also illustrates the concrete influence of AI on entrepreneurial decisions. If you are thinking about using AI in your company, please do not hesitate to contact us. We look forward to hearing from you and discussing your specific question and first steps!