DAMO Academy, Alibaba's research and innovation institute, today announced that its sensing algorithm has enabled high-beam simulation of low-wire-beam LiDAR. This indirectly increases the LiDAR wiring harness by a factor of three or more, enabling low-cost generic LiDAR to replace high-cost radar, it said.
LiDAR that can detect obstacles is the most important "eye" of an automated vehicle, and this algorithm breakthrough is equivalent to "using a low-resolution camera to capture a DSLR effect", which can significantly reduce the cost of automated driving perception components.
In automated driving scenarios, a high density of LiDAR is often required to meet perceptual requirements, and the high cost of LiDAR above 64 wires has become one of the bottlenecks for large-scale commercialization of automated driving.
The DAMO Academy Autopilot Lab environment perception algorithm combines camera images to perform deep complementary and semantic recognition of low-wire-beam LiDAR point clouds.
This results in a more dense 3D reconstruction of the LiDAR point cloud map, which not only reads the distance and shape of obstacles more accurately, but also determines their category information more precisely.
The top image is the original low-beam LiDAR point cloud, and the bottom image is the point cloud after deep completion by the DAMO Academy algorithm.
In terms of accuracy metrics, the DAMO Academy achieves the industry average of using high wire-beam LiDAR inputs, using low wire-beam LiDAR inputs.
According to the company, the average error in reading obstacle distance information within 50 meters is about 25 cm, while the DAMO Academy achieves 100fps processing power for depth completion tasks.