Load Inference Model for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving vehicles lack an effective method to dynamically adjust actuator commands based on load conditions without weigh sensors, which are not always available, and need to determine when loads reach maximum capacity.
Innovation Solution
A load inference system using a neural network regression model trained with data collected during manual driving sessions, where features such as speed, driving commands, and gear shifts are recorded to infer the load in real-time, allowing for dynamic adjustments of throttle and brake commands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If weigh sensors are installed to measure load conditions, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent replaces the mechanical weigh sensor system with a computational model that uses existing vehicle sensor data (accelerometer, GPS, vehicle parameters) to infer load conditions. The load inference model processes mechanical and electrical signals from existing sensors to estimate load weight, eliminating the need for dedicated weigh sensors while maintaining measurement capability.
Solution Approach 2:
The patent introduces a load inference model as an intermediary computational layer between existing vehicle sensors and the load measurement function. This model acts as a mediator that translates data from accelerometers, GPS, and vehicle parameter sensors into accurate load estimates without requiring direct weight measurement hardware.
2Reliability
If load measurement systems are added to determine maximum capacity, then reliability is improved, but device complexity increases
Solution Approach 1:
The patent makes the existing vehicle sensor system multi-functional by enabling it to serve both its original navigation and control functions plus the additional load measurement function. The same accelerometer, GPS, and vehicle parameter sensors used for autonomous driving also provide data for load inference, eliminating the need for dedicated load measurement hardware.
Solution Approach 2:
The vehicle's existing sensor system serves itself by providing data for both primary autonomous driving operations and secondary load measurement functions. The load inference model utilizes data already being collected by the vehicle's sensors, allowing the system to self-determine load capacity without external or additional measurement devices.
3Ease of operation
If real-time load monitoring is implemented without weigh sensors, then ease of operation is improved, but measurement precision may worsen
Solution Approach 1:
The patent implements a feedback mechanism where the load inference model continuously receives real-time data from vehicle sensors and dynamically adjusts actuator commands based on inferred load conditions. The system monitors vehicle acceleration, speed, and other parameters, processes this information through the load model, and uses the resulting load estimates to modify throttle and brake commands in real-time, improving operational ease while maintaining precision through continuous validation.
Data Source
AI summary
According to various embodiments, described herein are methods and systems for collecting data for training a load inference regression model for use in an ADV. According to one exemplary method, an ADV is manually driven on a segment of a road for a number of periods of time. During each period of time, a set of features of the ADV are recorded, including one or more features at a first time prior to a gear shift from a first gear position to a second gear position, and one or more features at a second time after the gear shift. For each of the number of periods of time, a weight of the ADV is also recorded using a weight sensor. The recorded features and the total weight of the ADV for each of the periods of time are then used to train a neural network regression model for inferring a load of the ADV in real time.