Waypoint Generation Using IMU and Vision for Smoother Vehicle Actuation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current assisted and autonomous driving systems face challenges in efficiently determining vehicle paths due to the complexity and resource-intensive nature of environmental data processing, necessitating a more efficient method for generating waypoints for smoother actuation.
Innovation Solution
A method using a graph neural network and kinematics equations to estimate waypoints from visual odometry and inertial measurement unit signals, training a model with only four waypoints spaced 0.5 seconds apart, and employing a computer-implemented system to generate and utilize these waypoints for smoother vehicle actuation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex environmental data processing is used to determine vehicle paths, then path determination accuracy is improved, but resource consumption increases
Solution Approach 1:
The path determination process is segmented into two stages: (1) a lightweight model generates initial waypoints using minimal sensor data, and (2) these waypoints are refined by a downstream controller. This segmentation allows the system to achieve accurate path determination while consuming fewer resources during the critical real-time decision-making phase.
Solution Approach 2:
The system performs partial action by generating only four key waypoints instead of computing complete continuous paths. These four waypoints capture the essential trajectory information needed for safe and accurate navigation, while avoiding the excessive computational resources required for full environmental processing and continuous path planning.
2Stability of the object's composition
If multiple waypoints are generated for path prediction, then actuation smoothness is improved, but computational complexity increases
Solution Approach 1:
The system generates exactly four waypoints spaced 0.5 seconds apart, which is sufficient to provide the downstream controller with adequate information for calculating smooth vehicle paths over a longer time span. This partial action approach achieves actuation smoothness without the excessive computational complexity of generating numerous waypoints.
3Measurement precision
If more waypoints are used for model training, then prediction accuracy is improved, but training resource requirements increase
Solution Approach 1:
The system uses only four waypoints for model training, which has been found to achieve high prediction accuracy. This partial approach significantly reduces the quantity of training data required compared to using numerous waypoints, while still providing sufficient information for the model to learn effective path prediction patterns.
Data Source
AI summary
A method for training of machine learning processes for autonomous driving applications, the method includes (a) receiving a training dataset comprising images acquired during a driving episode of a vehicle, the images associated with inertial measurement unit information; (b) estimating, based on the images and the inertial measurement unit information, waypoints indicative of waypoints passed by the vehicle during the driving episode; and (c) training a machine learning process using the images and the estimated waypoints, to produce a predictable set of future waypoints indicative of future waypoints on a driving route of a vehicle.