Self-supervised Velocity Learning for Autonomous Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional object detection and tracking systems using RADAR data and machine learning lag in determining object velocities due to delayed correlation of detections and reliance on time-based position tracking, leading to inaccuracies and prolonged determination times.
Innovation Solution
Deployment of machine learning models, such as neural networks, to estimate object velocities based on measured RADAR data, comparing estimated velocities with expected RADAR data to iteratively refine velocity determinations using multiple RADAR sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional time-based position tracking is used to determine object velocities, then the system can maintain simplicity in the detection approach, but the velocity determination accuracy deteriorates and determination time increases
Solution Approach 1:
The system performs preliminary correlation of RADAR detections with bounding boxes before velocity calculation, establishing detection-object associations in advance. This allows velocity to be computed directly from correlated detections without waiting for multiple tracking frames, reducing determination time while improving accuracy through self-supervised learning.
Solution Approach 2:
The patent replaces traditional mechanical time-based position tracking with a machine learning-based velocity estimation system. The ML model directly estimates velocity from RADAR data and bounding box correlations, substituting the mechanical tracking approach with an intelligent system that processes data more efficiently and accurately.
2Speed
If RADAR detections are correlated with bounding boxes only after accumulating threshold tracking information, then the system maintains simplicity in detection, but velocity determination lags behind other determinations
Solution Approach 1:
The system performs preliminary correlation of RADAR detections with bounding boxes before velocity calculation, establishing detection-object associations in advance. This allows velocity to be computed directly from correlated detections without waiting for multiple tracking frames, reducing determination time while improving accuracy through self-supervised learning.
Solution Approach 2:
The system uses self-supervised learning where the ML model learns to correlate detections with bounding boxes and estimate velocities autonomously from the data itself, without requiring external supervision or complex predefined rules. The model serves itself by learning patterns directly from RADAR data and tracking information.
Data Source
AI summary
Embodiments of the present disclosure relate to a system and method used to transfer image data via Ethernet. In some embodiments, the method may include determining, using a machine learning model, an estimated velocity corresponding to an object based at least on measured RADAR data, where the measured RADAR data may correspond to RADAR detections associated with the object. In some embodiments, the method may further include determining expected RADAR data corresponding to the object based at least on the estimated velocity. Some embodiments may additionally include updating one or more parameters of the machine learning model based on the difference between the measured RADAR data and the expected RADAR data.