Semantic Occupancy Maps for Autonomous Vehicle Trajectory Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in effectively planning trajectories that account for future occupancy of various semantic objects, such as vehicles and pedestrians, due to limitations in sensor data interpretation and map integration.
Innovation Solution
A computer-implemented method using a vehicle computing system that fuses LIDAR sensor data and map data to generate interpretable semantic occupancy maps, predicting the occupancy of specific object types at future time steps, and selecting trajectories based on these maps to ensure safe navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional sensor data interpretation methods are used for trajectory planning, then the system complexity is lower, but the measurement precision and reliability of predicting future occupancy are insufficient
Solution Approach 1:
The patent segments the occupancy prediction task by creating separate occupancy maps for different semantic object types (vehicles, pedestrians, cyclists, etc.). Each map independently predicts occupancy for its specific object type, allowing the system to achieve high prediction accuracy for each category while maintaining manageable system complexity through modular processing.
Solution Approach 2:
The patent introduces semantic occupancy maps as intermediary representations between raw sensor data and trajectory planning decisions. These maps serve as a structured intermediate format that captures future occupancy predictions with semantic meaning, enabling both high measurement precision and efficient processing by decoupling the complexity of interpretation from the final decision-making process.
2Reliability
If detailed semantic occupancy maps for multiple object types are generated, then the reliability and measurement precision improve, but the data storage requirements and processing cycles increase
Solution Approach 1:
The patent extracts only the essential occupancy information for each semantic object type into dedicated occupancy maps, rather than storing or processing complete raw sensor data. Each map contains only the relevant occupancy predictions for its specific object type, significantly reducing data storage requirements while maintaining the reliability needed for safe navigation decisions.
Solution Approach 2:
The system generates occupancy maps for multiple future time steps and multiple semantic object types, which appears excessive at first glance. However, this partial action approach focuses computational resources only on predicting occupancy (the critical safety parameter) rather than processing all sensor data comprehensively, achieving high reliability with optimized data quantities.
3Measurement precision
If multiple machine-learned models are used to determine occupancy for different semantic object types, then the measurement precision and reliability improve, but the use of energy and processing cycles increase
Solution Approach 1:
The patent divides the occupancy prediction task into multiple specialized machine-learned models, each dedicated to a specific semantic object type (vehicles, pedestrians, cyclists). This segmentation allows each model to be optimized for its specific target, achieving high measurement precision while improving energy efficiency by avoiding the need for a single complex model to process all object types equally.
Solution Approach 2:
While using multiple specialized models, the system achieves a form of universality where the collective set of models covers all relevant semantic object types for autonomous navigation. Each model is simple and energy-efficient for its specific function, but together they provide comprehensive occupancy detection accuracy across all object types, optimizing the balance between energy use and measurement precision.
Data Source
AI summary
Systems and methods for generating semantic occupancy maps are provided. In particular, a computing system can obtain map data for a geographic area and sensor data obtained by the autonomous vehicle. The computer system can identify feature data included in the map data and sensor data. The computer system can, for a respective semantic object type from a plurality of semantic object types, determine, by the computing system and using feature data as input to a respective machine-learned model from a plurality of machine-learned models, one or more occupancy maps for one or more timesteps in the future, and wherein the respective machine-learned model is trained to determine occupancy for the respective semantic object type. The computer system can select a trajectory for the autonomous vehicle based on a plurality of occupancy maps associated with the plurality of semantic object types.