An American multinational technology company that is a supplier of artificial intelligence hardware and software, who also designs system on chip units for the automotive market has won first place in a competition for autonomous driving development technology.
NVIDIA’s work that won the 3D Occupancy Prediction Challenge for autonomous driving development was presented at the Computer Vision and Pattern Recognition Conference held June 18-22 in Vancouver, Canada. NVIDIA also won the Innovation Award for “‘fresh insights into the development of view transformation modules,” with “substantially improved performance,’” a release said.
NVIDIA’s work introduced 3D occupancy prediction for safe self-driving systems, the release said.
“3D occupancy prediction is the process of forecasting the status of each voxel in a scene, that is, each data point on a 3D bird’s-eye-view grid,” the release said. “Voxels can be identified as free, occupied or unknown.”
3D occupancy grid prediction is “critical to the development of safe and robust self-driving systems,” the release said.
“NVIDIA’s winning solution features two important AV advancements,” Zhiding Yu, senior research scientist for learning and perception at NVIDIA, said in the release. “It demonstrates a state-of-the-art model design that yields excellent bird’s-eye-view perception. It also shows the effectiveness of visual foundation models with up to 1 billion parameters and large-scale pretraining in 3D occupancy prediction.”
Perception for autonomous driving has evolved from handling 2D tasks (detecting objects or free spaces in images) “to reasoning about the world in 3D with multiple input images.”
Jose Alvarez, director of AV applied research and distinguished scientist at NVIDIA, said in the release that reasoning in 3D gives “a flexible and precise fine-grained representation of objects in complex traffic scenes and is “critical for achieving the safety perception requirements for autonomous driving.”
Traditional 3D object detection is limited in that it lacks expressiveness in which the 3D bounding boxes that detect and represent objects in a scene may “not represent enough real-world information,” the release said.
“It also requires defining taxonomies and ground truths for all possible objects, even ones rarely seen in the real world, such as road hazards that may have fallen off a truck,” the release said. “In contrast, 3D occupancy prediction provides rich information about the world to a self-driving vehicle’s planning stack, which is necessary for end-to-end autonomous driving. Software-defined vehicles can be continuously upgraded with new developments that are proven and validated over time.”
