On-Board Machine Learning Model for Autonomous Vehicle Motion Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately predicting the future locations of objects in their environment, which hinders their ability to generate effective motion plans and ensure safety.
Innovation Solution
A computer-implemented method using a machine-learned model on-board the autonomous vehicle to determine vehicle actions based on features extracted from state data of proximate objects, such as pedestrians, vehicles, or bicycles, allowing the vehicle to decide whether to stop, pass, or adjust speed accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional motion control methods are used, then the system is simpler to implement, but the accuracy in predicting future locations of objects deteriorates
Solution Approach 1:
The patent replaces traditional mechanical motion control systems with a machine-learned model that processes sensor data to predict object locations and determine vehicle actions. This substitution enables higher prediction accuracy through learned patterns while managing complexity through integrated on-board processing.
Solution Approach 2:
The system changes the parameters of motion control by using machine learning to dynamically adjust vehicle actions based on predicted object locations. The model processes multiple features (position, velocity, acceleration) and transforms them into optimized control decisions, improving prediction accuracy through parameter optimization.
2Measurement precision
If more processing power is allocated to improve prediction accuracy, then the prediction quality improves, but the memory usage and computational burden increase
Solution Approach 1:
The patent segments the motion control problem into distinct processing stages: sensor data acquisition, feature extraction from state data, machine learning inference, and action determination. This segmentation allows efficient memory management by processing data in stages rather than loading all data simultaneously, reducing overall memory requirements while maintaining prediction accuracy.
Solution Approach 2:
The system extracts only the essential features from raw state data that are necessary for prediction, rather than processing all available sensor data. This extraction principle reduces the data volume requiring memory storage and processing while preserving the critical information needed for accurate predictions.
3Loss of time
If a machine-learned model is implemented on-board, then the response time is reduced and adaptability improves, but the device complexity increases
Solution Approach 1:
The machine-learned model is pre-trained offline with extensive data before deployment in the autonomous vehicle. This preliminary training action transfers learned knowledge to the on-board system, enabling fast real-time inference without requiring complex computational resources during actual operation, thus reducing response time while managing on-board complexity.
Solution Approach 2:
The patent implements a copy of the trained machine-learned model on the on-board computing system, allowing the vehicle to independently perform predictions without continuous external processing. This copying enables fast local decision-making while the original model training can be performed separately with more resources.
Data Source
AI summary
Systems and methods for controlling the motion of an autonomous are provided. In one example embodiment, a method includes obtaining, by one or more computing devices on-board an autonomous vehicle, state data descriptive of a current state of an object that is proximate to the autonomous vehicle. The method includes determining, by the one or more computing devices, one or more features associated with the object based at least in part on the state data. The method includes determining, by the one or more computing devise, a vehicle action for the autonomous vehicle based at least in part on a machine-learned model and the one or more features associated with the object. The vehicle action includes stopping a motion of the autonomous vehicle or moving the autonomous vehicle past the object.