Vehicle Control Using Unified Sensor Messages and Dual-Cycle Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle control systems rely solely on local data models, which are limited in information processing capability and sensor data utilization, failing to effectively manage complex driving scenarios.
Innovation Solution
A vehicle control method that transforms sensor information of different modalities into unified modality short-cycle message information, transmitting it to a remote data model for real-time processing, and generating long-cycle message information to trigger more accurate vehicle control decisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If local data models are used for vehicle control, then system complexity is reduced, but information processing capability is limited
Solution Approach 1:
The system divides control tasks into two segments: real-time control tasks executed by local data models on the vehicle, and complex scenario analysis tasks executed by remote data models in the cloud. This segmentation allows each component to operate within its capability boundaries while collectively achieving superior performance.
Solution Approach 2:
A communication interface acts as an intermediary between local and remote data models, transmitting sensor information and control instructions. This intermediary enables the local system to leverage remote computing power without directly integrating complex remote infrastructure into the vehicle.
2Loss of information
If remote data models are used for vehicle control, then information processing capability is improved, but real-time performance deteriorates
Solution Approach 1:
Control tasks are segmented by time sensitivity: real-time tasks (braking, acceleration) are handled locally without remote communication delay, while non-time-critical tasks (complex scenario analysis) are offloaded to remote models.
Solution Approach 2:
The system performs preliminary processing of sensor data locally to extract only essential information needed by remote models, reducing communication overhead and enabling faster remote processing responses.
3Loss of time
If all sensor information is processed locally, then real-time performance is maintained, but intelligent degree deteriorates
Solution Approach 1:
The system extracts only the most critical sensor information needed for remote analysis, leaving detailed real-time processing to local models. This extraction enables remote models to provide enhanced intelligence without overwhelming communication bandwidth.
Solution Approach 2:
The system adds a remote processing dimension to the traditional local-only architecture, creating a hierarchical structure where local and remote models operate at different levels of abstraction and time scales.
4Speed
If local systems are used for vehicle control, then response speed is improved, but adaptability to complex scenarios deteriorates
Solution Approach 1:
Scenarios are segmented into simple routine scenarios handled by local models and complex novel scenarios referred to remote models for analysis, optimizing both response speed and adaptability.
Solution Approach 2:
The system implements feedback loops where remote model results are used to update and improve local model performance over time, enabling local models to gradually handle more complex scenarios with fast response.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A vehicle control method includes obtaining sensor information of different modalities of different onboard devices of a vehicle, performing modality transformation on the sensor information to generate short-cycle message information of a unified modality, sending the short-cycle message information to a first data model at a remote end, generating long-cycle message information by a second data model configured at the vehicle based on the sensor information with the different modalities, sending the long-cycle message information to the first data model at the remote end, receiving the vehicle control information returned by the first data model at the remote end, and controlling the vehicle based on the vehicle control information. The short-cycle message information has better real-time performance than the long-cycle message information, and the long-cycle message information is used to trigger the first data model to generate vehicle control information.