Cut-In Probability Prediction Using Vehicle Trajectory Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for predicting cut-ins by autonomous vehicles rely on inferred predictions from predicted future paths, which can be inaccurate and fail to capture complex behaviors or non-linear interactions, leading to potential collisions.
Innovation Solution
A cut-in neural network trained directly on real-world driving examples generates probabilities of surrounding agents cutting in front of the vehicle, using trajectory data represented as 2D channels processed by a convolutional neural network to improve prediction accuracy and resource efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If inferred predictions from predicted future paths are used to predict cut-ins, then the prediction process is simpler, but the prediction accuracy deteriorates and complex behaviors are not captured
Solution Approach 1:
The patent replaces traditional mechanical inference methods with a neural network-based system. Instead of using simple geometric or kinematic reasoning to predict cut-ins, the system employs a neural network that processes trajectory data and sensor information to generate predictions, thereby improving accuracy while maintaining computational efficiency
Solution Approach 2:
The patent changes the input parameters from simple predicted future paths to comprehensive trajectory data including historical positions, velocities, and sensor readings. This parameter transformation allows the neural network to capture complex behavioral patterns and improve prediction accuracy without significantly increasing system complexity
2Use of energy by moving object
If traditional prediction methods are used, then computational resources are saved, but prediction accuracy and ability to capture non-linear interactions deteriorates
Solution Approach 1:
The patent replaces computationally intensive traditional prediction algorithms with a neural network that has been pre-trained offline. The network performs efficient inference during runtime, capturing non-linear interactions and improving reliability without consuming excessive computational resources during vehicle operation
Solution Approach 2:
The neural network is trained in advance using extensive simulation data and real-world trajectory information. This preliminary training phase allows the network to learn complex patterns offline, so that during actual vehicle operation, predictions can be made quickly and efficiently with minimal real-time computational overhead
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating cut-in probabilities of agents surrounding a vehicle. One of the methods includes obtaining agent trajectory data for one or more agents in an environment; obtaining vehicle trajectory data of a vehicle in the environment; and processing a network input generated from the agent trajectory data and vehicle trajectory data using a neural network to generate a cut-in output, wherein the cut-in output comprises respective cut-in probabilities for each of a plurality of locations in the environment, wherein the respective cut-in probability for each location that is a current location of one of the one or more agents characterizes a likelihood that the agent in the current location will intersect with a planned future location of the vehicle within a predetermined amount of time.