MOE Neural Network for Joint Autonomous Driving Tasks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network-based autonomous driving solutions fail to consider the mutual impact and requirements of prediction, decision, and planning, leading to results that do not meet actual driving requirements, and lack generalization capability in diverse traffic environments.
Innovation Solution
A neural network architecture using a mixture of experts (MOE) network with parallel head networks for feature extraction and fusion, enabling independent optimization of each task while maintaining interaction, and incorporating self-attention and cross-attention networks for enhanced feature extraction and fusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single unified neural network is used for prediction, decision, and planning, then model interaction is enhanced, but task independence and optimization flexibility are reduced
Solution Approach 1:
The neural network is segmented into a shared backbone network and multiple independent head networks, where each head network corresponds to a specific autonomous driving task (prediction, decision, planning). This segmentation allows each task to be independently optimized while sharing common feature extraction capabilities from the backbone, resolving the contradiction between model interaction and task independence.
2Adaptability or versatility
If multiple independent neural networks are used for each autonomous driving task, then task independence is maintained, but computing power consumption and latency increase
Solution Approach 1:
Multiple independent head networks are merged into a unified neural network architecture that shares a common backbone network for feature extraction. This merging reduces redundant computing operations and lowers overall computing power consumption and latency, while each head network maintains its independence for task-specific optimization.
3Ease of manufacture
If existing neural network solutions are used without considering mutual impact of tasks, then implementation is simpler, but accuracy and generalization capability are reduced
Solution Approach 1:
The neural network incorporates feedback mechanisms where the shared backbone network receives combined loss signals from multiple head networks during training. This feedback allows the model to learn from the mutual impact of different tasks (prediction, decision, planning), improving overall accuracy and generalization capability while maintaining a relatively simple unified architecture.
Data Source
AI summary
A neural network, and an autonomous driving method and apparatus are provided. The neural network includes a mixture of experts MOE network and N parallel head networks. The MOE network is used to perform feature extraction and feature fusion on input data to obtain N types of feature data, where the input data includes ego vehicle data, obstacle data, road topology data, and navigation data, the N types of feature data are in a one-to-one correspondence with the N parallel head networks, and each of the N parallel head networks corresponds to one autonomous driving task; and each head network performs one corresponding autonomous driving task based on one type of feature data that is correspondingly input, where the autonomous driving task includes a prediction task, a decision task, and a planning task. Each autonomous driving task can maintain independence to some extent while maintaining certain interaction.