Multi-Task Intent Modeling for Faster Autonomous Object Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately detecting and predicting the behavior of objects in their surroundings, which can impact safety and navigation efficiency.
Innovation Solution
A computing system equipped with a multi-task machine-learned intent model that processes sensor data and map data to jointly determine object detection, trajectory forecasting, and behavior intention in a single forward pass, utilizing shared layers for feature determination across tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate machine-learned models are used for object detection, trajectory forecasting, and behavior intention determination, then each task can be optimized independently, but the processing time and computational resources increase significantly
Solution Approach 1:
The patent combines multiple separate machine-learned models (object detection, trajectory forecasting, behavior intention determination) into a single unified multi-task model. This model processes sensor data and map data simultaneously to produce all three outputs (detection, trajectory, behavior) in a single forward pass, thereby reducing processing time while maintaining accuracy through shared feature extraction layers.
Solution Approach 2:
The unified machine-learned model is designed to perform multiple functions simultaneously: it detects objects, forecasts their trajectories, and determines their behavior intentions all within a single model architecture. This multi-functional approach eliminates the need for separate specialized models while optimizing resource utilization and processing efficiency.
2Measurement precision
If multiple separate models are deployed for different tasks, then task-specific accuracy can be maximized, but the computational resources and processing overhead increase
Solution Approach 1:
The patent merges multiple task-specific models into a single multi-task model that handles object detection, trajectory forecasting, and behavior intention determination concurrently. This consolidation reduces computational overhead and improves processing efficiency while maintaining prediction accuracy through shared feature extraction and task-specific output layers.
Solution Approach 2:
While unified, the model employs segmentation of functions through shared layers for common feature extraction and separate head networks for task-specific predictions. This architectural segmentation allows efficient resource sharing while preserving task-specific optimization capabilities.
3Loss of information
If comprehensive sensor data and map data are processed through multiple separate models, then complete object understanding is achieved, but the system complexity and processing time increase
Solution Approach 1:
The patent combines multiple processing models into a single unified system that simultaneously processes sensor data and map data to produce comprehensive object understanding including detection, trajectory, and behavior information. This merging reduces system complexity by eliminating multiple separate model deployments while maintaining complete object understanding through integrated multi-task processing.
Data Source
AI summary
Generally, the disclosed systems and methods utilize multi-task machine-learned models for object intention determination in autonomous driving applications. For example, a computing system can receive sensor data obtained relative to an autonomous vehicle and map data associated with a surrounding geographic environment of the autonomous vehicle. The sensor data and map data can be provided as input to a machine-learned intent model. The computing system can receive a jointly determined prediction from the machine-learned intent model for multiple outputs including at least one detection output indicative of one or more objects detected within the surrounding environment of the autonomous vehicle, a first corresponding forecasting output descriptive of a trajectory indicative of an expected path of the one or more objects towards a goal location, and/or a second corresponding forecasting output descriptive of a discrete behavior intention determined from a predefined group of possible behavior intentions.