Vehicle Trajectory Planning Using Temporal Classification and RNN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current path and trajectory planning methods for motor vehicles face challenges in robustness and calculation complexity, particularly with indirect approaches like the Werling method and high complexity algorithms like the Mercy method, which lead to inefficient resource usage and potential failure in generating optimal solutions.
Innovation Solution
The implementation of a device using a temporal classification means with compact representations and a recurrent neural network (RNN-LSTM) for planning paths and trajectories, shifting calculation resource requirements from the online inference phase to an offline training phase, allowing all input parameters to be considered while reducing complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If indirect approach (Werling method) is used to solve optimal control problem, then some input parameters can be selected and problem solved partially, but the use lacks robustness owing to large number of input parameters and many calculations are rejected due to collision risk
Solution Approach 1:
The patent applies preliminary action by performing temporal classification and compact representation of input parameter sequences during an offline training phase before the actual path planning is needed. This pre-processing creates a structured database that can be quickly queried during online operation, avoiding the need to handle large numbers of input parameters in real-time while ensuring robustness through pre-validated collision avoidance constraints.
2Reliability
If direct approach (Mercy method) is used to solve optimal control problem, then all constraints are considered in solving, but the complexity of algorithm makes it impossible to incorporate into device on board motor vehicle
Solution Approach 1:
The patent segments the path planning problem into two distinct phases: an offline training phase where temporal classification and compact representation are performed to build a structured database, and an online inference phase where pre-computed patterns are quickly matched to current situations. This segmentation reduces online algorithm complexity while maintaining comprehensive constraint satisfaction achieved through the thorough offline analysis.
Solution Approach 2:
The patent applies preliminary action by performing temporal classification and compact representation of input parameter sequences during an offline training phase before the actual path planning is needed. This pre-processing creates a structured database that can be quickly queried during online operation, avoiding the need to handle large numbers of input parameters in real-time while ensuring robustness through pre-validated collision avoidance constraints.
3Ease of operation
If potential field methods are used, then obstacles generate anti-gravity field to prevent vehicle approach, but methods entail risk of blocking in local minimum and may fail to generate best solution
Solution Approach 1:
The patent replaces the mechanical potential field approach with a data-driven temporal classification system using compact representations. Instead of using anti-gravity fields that can trap vehicles in local minima, the system classifies sequences of input parameters temporally and retrieves pre-computed optimal paths from a database, guaranteeing optimal solutions without the mathematical limitations of potential field methods.
Data Source
AI summary
A device for planning a path and/or a trajectory for a motor vehicle includes a module for receiving a sequence of input variables and hardware and software for determining a control law corresponding to the path and/or the trajectory according to the sequence of input variables received. The device includes a temporal classification unit in a set of compact representations.