Vehicle Sensor Fusion Occupancy Grids for Trajectory Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Secondary and low-level vehicle systems are constrained by limited computing resources and input data, leading to inaccurate object detection and prediction due to the strengths and weaknesses of different sensor modalities like lidar and radar, resulting in false positives and negatives.
Innovation Solution
Fusing lidar and radar data to create feature maps that identify non-drivable, over-drivable, and under-drivable objects using a machine-learned model to generate an occupancy grid, which validates vehicle trajectories and reduces false positives and negatives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data from different modalities (lidar, radar) is used for object detection in low-level systems, then detection coverage is improved, but false positives and false negatives increase due to the strengths and weaknesses of each sensor modality
Solution Approach 1:
The patent combines sensor data from multiple modalities (lidar, radar, cameras) into a unified occupancy grid representation. By merging the strengths of different sensors - lidar for precise 3D spatial information, radar for velocity and all-weather capability, cameras for semantic information - the system achieves more reliable object detection while reducing false positives and negatives that occur when using individual sensor modalities alone.
2Extent of automation
If secondary and low-level systems perform object detection and trajectory validation, then system autonomy is improved, but computing resource constraints limit the complexity of processing and analysis
Solution Approach 1:
The patent divides the autonomous driving system into distinct hierarchical levels: primary systems for high-level planning and secondary/low-level systems for real-time perception and trajectory validation. This segmentation allows each level to operate within its computing resource constraints while maintaining overall system autonomy. The occupancy grid approach further segments the environment into discrete cells, enabling efficient processing on resource-constrained low-level systems.
Solution Approach 2:
The patent creates a simplified computational representation (occupancy grid) that copies essential spatial and semantic information from complex multi-sensor data. This abstracted representation allows low-level systems to perform trajectory validation without processing the full complexity of raw sensor streams, effectively reducing computational requirements while preserving critical detection capabilities.
3Measurement precision
If lidar data is used for elevation measurements, then precision for identifying over-drivable and under-drivable objects is improved, but performance in inclement weather deteriorates
Solution Approach 1:
The patent creates a composite sensing system that integrates multiple sensor modalities, analogous to composite materials combining different properties. By fusing lidar data (providing precise elevation measurements for identifying over-drivable and under-drivable objects) with radar data (providing all-weather capability and velocity information), the system maintains measurement precision while achieving reliability in inclement weather conditions where lidar alone would fail.
Data Source
AI summary
Techniques for fusing sensor data generated by different sensor modalities to improve object detections and object predictions determined by low-level systems of a vehicle. The techniques may include determining feature maps based on sensor data generated by different sensor modalities associated with a vehicle. In some examples, the feature maps may include at least a first feature map indicative of a location of an object in an environment of the vehicle and a second feature map indicative of elevation information associated with the object. The techniques may also include inputting the first feature map and the second feature map into a machine-learned model associated with the low-level system of the vehicle. In some examples, an output may be received from the machine-learned model that includes an occupancy grid, and the occupancy grid may exclude representation(s) associated with over-drivable object(s) and/or under-drivable object(s) that may be disposed in the environment.