Various sensors, such as LiDAR, cameras and radar, installed on the vehicle can collect environment data surrounding the vehicle. Using sensor fusion technology, perception algorithms running on the vehicle can determine in real time the type, location, as well as the velocity and orientation of objects on the road. Supporting this autonomous perception system is Baidu’s big data and deep learning technology developed over the years along with a vast collection of real-world, labelled driving data. A large-scale, deep learning platform and GPU clusters drastically shorten learning time for large quantities of data, and the newly trained models are deployed onto the vehicle using over-the-air updates through the cloud. Artificial intelligence + data-driven solutions enable Apollo’s perception system to continuously improve its detection and recognition capabilities in order to provide accurate, stable, and reliable input for other autonomous system modules.
As a key component of Apollo, simulation provides a crucial service by virtually driving millions of kilometers on daily basis using Apollo’s vast collection of real-world traffic and autonomous driving scene data.By using Apollo’s open simulation service, partners gain access to a large number of autonomous driving scenes to quickly test, validate, and optimize models with comprehensive coverage that is safe and efficient.
Baidu pioneered the extensive application of deep learning and artificial intelligence technology for map creation and is one of the few Chinese firms capable of producing HD mapping data on a large scale. Based on GPS, IMU, HD map, and a variety of sensor input, Apollo’s localization system is a comprehensive positioning solution that supports down to centimeter-level accuracy. Based on different usage scenarios, the integrated product can be customized with software and hardware while minimizing costs and having adjustable precision.
End-To-End autonomous solutions are being explored because of its low cost and low engineering complexity. By using the large number of real road data collected by the map acquisition vehicle, the horizontal and latitude driving models were constructed based entirely on the depth learning, and the practice was carried out quickly on the real car. We open horizontal, latitude model source code and 10,000 km of data.
The vehicle is equipped with prediction, behavior/motion planning systems. Apollo’s system can intelligently optimize the driving plan according to real-time traffic conditions, speed limit, etc. and get precise trajectories that are both safe and comfortable. Apollo can provide capability of all time auto cruising / vehicle following on fixed route.
Apollo’s intelligent vehicle control and canbus-proxy modules are more precious , generic and adaptive to different environment. The modules can handle different road conditions, different speed, different vehicle types and canbus protocols.Apollo provides waypoint following autonomous driving capability with control accuracy of ~10 cm.
By opening up autonomous driving source code, capabilities, and data, Apollo’s open data platform will form a comprehensive "vehicle + cloud" open ecosystem. Developers and partners with strong software development and AI research capabilities that lack the data and computing power can tap into a diverse array of fast and flexible services for driving data, computing power, and labeling features. By opening up relevant technologies and resources and pooling together developers and industry partners, we hope to build an open autonomous driving ecosystem that will empower each participant and contribute to the widespread adoption of autonomous driving.
Hardware is an essential part of autonomous driving. Apollo provides the global developers with complete hardware reference design, including the selection of vehicle, key hardware components, and peripherals. We also provide a detailed hardware installation guide. This guide elaborates the process of hardware installation and offers a starting point for hardware/software integration and vehicle road-testing.
Map Engine is a high-precision map data management service of vehicle terminal. It encapsulates the management mechanism of HDMap data, shields the details of the data and provides a unified data query interface to the application layer. It includes the core capabilities such as element retrieval, spatial retrieval, format adaptation and cache management etc and provides a modular, hierarchical, cross-platform, highly customizable, flexible and efficient programming interface that allows users to easily build a dedicated terminal map solution.
DuerOS provides a complete voice-based vehicle interaction solution and is committed to providing users with a one-stop-shop for in-vehicle services, such as navigation, virtual Q&A assistant, personalized audio content recommendations, etc. An open and consistent platform empowers the car industry and Apollo partners to strengthen in-vehicle services with rich and engaging experiences.
Apollo provides a complete security framework and relevant components in an isolated and trusted security system. It protects the system network using an on-vehicle firewall that separates the inside and outside network of the vehicle and ensures the network integrity of each subsystem. The firewall can scan and filter each instruction sent to the vehicle to ensure that only trustworthy commands are executed. As a deeply embedded kernel network security module, Apollo’s security system provides source verification, content encryption, and trust assessment for every piece of transmitted information. Starting alongside system bootup, it assesses and monitors each operation and feature in the autonomous driving system to ensure its’ legimatacy and trustworthiness. The Apollo security system also provides a complete and secure over-the-air (OTA) functionality to guard against hacking incidents during the system update process. By safeguaring network, OS, cloud, and OTA security, each and every Apollo component operates in a secure and orderly manner.