Environment Maps
José Gomes
Disrupting the automotive industry??| SW Integration Leader | Project Management | Safety | System verification | System automation
In my previous post, I highlighted the importance of perception in autonomous driving. From a high-level perspective, there are three key modules related to perception:
Today, I’ll dive into the third module: Environment Maps.
Any movement first requires a clear understanding of the environment. While the environment perception module focuses on identifying the actors within the environment, the environment map module is responsible for creating different representations of the surroundings to inform motion planning.
There are three main types of maps used in autonomous driving:
Occupancy Grid
An occupancy grid maps all static objects in the environment surrounding the vehicle. The environment is represented as a set of grid cells, with each cell assigned a probability of being occupied. This probabilistic approach allows the system to handle measurement uncertainty and improve the map over time.
Occupancy grids are primarily built using data from LiDAR, radar, ultrasonic sensors, and cameras. However, they pose challenges such as the need for significant computational power to process high-resolution data in real time, the influence of external factors on sensors, and the dynamic nature of environments where multiple moving objects must be accounted for.
Localization Map
The localization map is used by the localization module to improve ego-state estimation. These maps can be created in various ways, from complex real-time approaches like SLAM (Simultaneous Localization and Mapping) to pre-mapped environments where maps are created by specialized vehicles and stored onboard the autonomous vehicle. When combined with sensor data, these maps enable precise ego localization.
A key challenge is maintaining the accuracy of these maps as environments evolve and change over time.
Detailed Road Map
This map focuses on road components, including:
Detailed road maps are typically a combination of pre-recorded data and real-time information gathered by the perception module. Challenges here include the need for substantial computational power to process data in real-time, ongoing map maintenance, and effective sensor integration.
In Summary:
Environment perception is as important as it is complex. Different types of maps are essential, each extracting unique information from the vehicle’s sensors. These maps must be combined and fed into the motion planning module to ensure safe and efficient autonomous driving.