Making Bertha See Even More: Radar Contribution

Abstract

For decades, radar has been applied extensively in warfare, earth observation, rain detection, and industrial applications. All those areas are characterized by requirements such as high quality of service, reliability, robustness in harsh environment and short update time for environmental perception, and even imaging tasks. In the vehicle safety and driver assistance field, radars have found widespread application globally in nearly all vehicle brands. With the market introduction of the 2014 Mercedes-Benz S-Class vehicle equipped with six radar sensors covering the vehicles environment 360 • in the near (up to 40 m) and far range (up to 200 m), autonomous driving has become a reality even in low-speed highway scenarios. A large azimuth field of view, multimodality and a high update rate have been the key innovations on the radar side. One major step toward autonomous driving was made in August 2013. A Mercedes-Benz research S-Class vehicle-referred to at Mercedes as Bertha-drove completely autonomously for about 100 km from Mannheim to Pforzheim, Germany. It followed the well-known historic Bertha Benz Memorial Route. This was done on the basis of one stereo vision system, comprising several long and short range radar sensors. These radars have been modified in Doppler resolution and dramatically improved in their perception capabilities. The new algorithms consider that urban scenarios are characterized by significantly shorter reaction and observation times, shorter mean free distances, a 360 • interaction zone, and a large variety of object types to be considered. This paper describes the main challenges that Daimler radar researchers faced and their solutions to make Bertha see.INDEX TERMS Radar, automotive radar, autonomous driving.

Tracking Multiple Vehicles Using a Variational Radar Model

“…To identify the different objects and to extract trajectories over time, each state vector is augmented with a unique label ∈ L from the label space L. This yields the labeled state vector x k = [x T k , ] T . All present vehicles are combined in the multi-object state which is modeled as the random finite set 3 (RFS) X k = {x (1) k , . .…”

“…In each measurement cycle, a radar sensor provides a set of detections Z k = {z (1) k , ..., z (m) k } ⊂ Z from the measurement space Z which either originate from actual vehicles, sensor noise, or other objects that are not relevant to the vehicle tracking task. Their number m may change from cycle to cycle.…”

“…3. In particular, a top view 1 The variational radar model is available at https://github.com/A-Scheel/Variational-Radar-Model Fig. 3b.…”

Why did the robot cross the road? — Learning from multi-modal sensor data for autonomous road crossing

“…Recently, approaches have been developed that tackle the problem independent of the type of intersection and without the use of vehicle-to-vehicle communication systems [8,19]. Dickmann et al augment their vehicle setup with additional radar sensors mounted to the left and right, to enable information processing in roundabout and intersection scenarios.…”

Radar Sensors for Autonomous Driving: Modulation Schemes and Interference Mitigation

“…For the operation in urban areas, vulnerable road users, such as pedestrians or cyclists, must be detected and classified [18]. As the reflected power from those road users is much lower than from vehicles, the radar sensor must be sensitive for weak targets [19].…”