Magna's Vision Transformer Drives 'Anytime, Anywhere' AD/ADAS
Magna, the company behind the 'anytime, anywhere' AD/ADAS concept, is pushing boundaries in autonomous driving. Their latest developments focus on cars that can make precise decisions without relying on traditional sensor inputs. The company's website uses advanced technologies like the Vision Transformer (ViT) model for spatial recognition and image classification, aiming for reliable autonomous driving in all conditions and without maps.
The ViT model, unlike Convolutional Neural Networks (CNNs), lacks local inductive bias but offers high generalizability. It requires substantial data to learn but improves performance with more training. ViT can detect people and vehicles based on visual information and surrounding scene structure, even in challenging conditions like nighttime or poor weather. The website's AD/ADAS architecture simplifies external environment recognition, route planning, and control using a learning approach.
To enhance ViT's performance, the website employs vehicle verification and simulators. Neural-net type planning technology is used to find low-risk routes based on data-driven models of surrounding dangers on regular roads. The ultimate goal is to achieve autonomous driving that works reliably in all conditions, without the need for maps.
Magna's 'anytime, anywhere' AD/ADAS concept, powered by the Vision Transformer model, is set to revolutionize autonomous driving. By focusing on accurate decision-making without sensor input and reliable navigation in all conditions, the company is paving the way for a society built around intelligent, self-driving cars.
