Nullmax Upgrades BEV-AI Architecture of Self-Driving to Build End-to-End Solution for High-Level Autonomy

Nullmax, a leading provider of ADAS/AD solutions and a pioneer in artificial intelligence, announces a major upgrade to its BEV-AI autonomous driving technology architecture. The new architecture integrates learning-based planning methods and perfects its algorithms, toolchains, and infrastructure. This upgrade enables the delivery of a comprehensive end-to-end AI system capable of achieving Level 3 and higher autonomous driving capabilities.

Introduced to autonomous driving perception in 2020, Nullmax’s BEV-AI architecture was first developed to fuse various perception data from an unlimited number and variety of sensors and continued to iterate and evolve over the past years. Now, Nullmax has achieved a substantial upgrade to the architecture to facilitate an end-to-end AI model, enabling high-level and human-like autonomous driving.

The innovative BEV-AI architecture allows the system to directly output perception results in BEV (Bird's Eye View) space and simultaneously make decisions and planning by a powerful, unified AI model. All subtasks of self-driving utilize learning-based algorithms instead of traditional rule-based algorithms, enabling the entire system to closely mimic human drivers.

This end-to-end system can understand the physical world and driving behaviors, thereby responding to various situations and planning driving trajectories like a human driver. Most importantly, the system boasts significant adaptability, swiftly expanding to any road, including unfamiliar scenarios never previously encountered.

Additionally, the new architecture incorporates cutting-edge algorithms and tools to support the development and deployment of the end-to-end system. These include high-performance, robust object detection algorithms, efficient automated data labeling, and unsupervised testing and evaluation tools. All these enhancements make the development and application of autonomous driving systems faster and more cost-effective.

Notably, Nullmax's BEV-AI combines CNN and Transformer methods, offering advanced autonomous driving features in high-end configurations and reliable ADAS functionalities in all setups. Thus, whether for complex or simple applications, all can benefit from the upgraded BEV-AI.

Learning-based approaches hold greater potential over rule-based methods. It’s more feasible with an end-to-end approach to realize L3 and L4 autonomous driving solutions that offer smarter functionalities. BEV-AI encompasses all technologies required for an end-to-end solution, including the collection, filtering, and labeling of data, the design, training, and evaluation of models, and the deployment, iteration, and testing of systems, enabling the AI model to learn real driving skills from vast amounts of data.

Nullmax has consistently explored the forefront of autonomous driving through its Machine Learning-First (MLF) technology strategy, utilizing AI, deep learning, and computer vision technologies. Committed to innovation, Nullmax continues to advance towards a driverless future.