ADAS & Autonomous Vehicles Special Report - December 2024 - 6

ARE TODAY'S SENSORS READY FOR
next-level automated
driving?
The Mercedes-Benz EQS sedan can use Drive Pilot
on major freeways in California and parts of Nevada.
A look at who's doing what when it comes to sensors for an L3 world.
by John Dinkel
S
AE Level 3 automated driving marks a clear break from the
lower levels of driving assistance since that is the dividing
line where the driver can be freed to focus on other things.
While the driver may sometimes be required to take control
again, responsibility in an accident can be shifted from the driver to
the automaker and suppliers. Only a few cars have met regulatory
approval for Level 3 operation. Thus far, only Honda (in Japan),
the Mercedes-Benz S-Class and EQS sedans with Drive Pilot and
BMW's recently introduced 7 Series offer Level 3 autonomy.
With more vehicles getting L3 technology and further automated
driving skills being developed, we wanted to check in with some
of the key players in this tech space and hear the latest industry
thinking about best practices for ADAS and AV Sensors.
Towards More Accurate 3D Object Detection for Robots
and Self-Driving Cars
Researchers from Japan's Ritsumeikan University have developed
DPPFA-Net, an innovative network that combines 3D LiDAR and
2D image data to improve 3D object detection for robots and
self-driving cars. Led by Professor Hiroyuki Tomiyama, the team
addressed challenges in accurately detecting small objects and
aligning 2D and 3D data, especially in adverse weather conditions.
DPPFA-Net incorporates three key modules:
* Memory-based Point-Pixel Fusion (MPPF):
Enhances robustness against 3D point cloud noise
by using 2D images as a memory bank.
* Deformable Point-Pixel Fusion (DPPF): Focuses on key pixel
positions for efficient high-resolution feature fusion.
6 DECEMBER 2024
* Semantic Alignment Evaluator (SAE): Ensures semantic
alignment between data representations during fusion.
The network outperformed existing models in the KITTI
Vision Benchmark, achieving up to 7.18% improvement in
average precision under various noise conditions. It also
performed well in a new dataset with simulated rainfall.
Ritsumeikan University researchers said this
advancement has significant implications for self-driving
cars and robotics. It could lead to reduced accidents,
improved traffic flow and safety, and enhanced robot
capabilities in various applications. The improvements
in 3D object detection are expected to contribute to
safer transportation, enhanced robot capabilities, and
accelerated development of autonomous systems.
AEVA
Aeva has introduced Atlas, the first 4D lidar
sensor designed for mass-production automotive
applications. Atlas aims to enhance advanced
driver assistance systems (ADAS) and autonomous
driving, meeting automotive-grade requirements.
The company's sensor is powered by two key
innovations: the fourth-generation lidar-on-chip
module called Aeva CoreVision that incorporate
all key lidar elements in a smaller package, using
silicon photonics technology, and the new Aeva
X1 new system-on-chip (SoC) lidar processor that
ADAS & AUTONOMOUS VEHICLES SPECIAL REPORT
MERCEDES-BENZ
ADAS & Autonomous Vehicles Special Report - December 2024

Table of Contents for the Digital Edition of ADAS & Autonomous Vehicles Special Report - December 2024
