Autonomous Driving Path Prediction in Adversity Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving systems face challenges in navigating through adversity conditions such as road work zones, obscured road markings, and obstacles due to their logic-based and cost-based decision-making approaches, which may not account for all scenarios, leading to potential errors.
Innovation Solution
Training machine learning models using vehicle sensor data, including image and inertial measurement data, to predict and prescribe safe navigation trajectories and detect adversity conditions, employing models like neural networks, random forests, or Naïve Bayes classifiers to provide control instructions for safe vehicle navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If logic-based and cost-based decision-making approaches are used in autonomous driving systems, then the systems can operate with defined rules and cost functions, but they fail to account for all possible scenarios and conditions leading to errors in tricky situations
Solution Approach 1:
The patent replaces logic-based and cost-based decision-making systems with machine learning models that learn navigation policies from training data. The ML models process sensor data (images, inertial measurements) and directly output navigation decisions, substituting the mechanical rule-based system with a data-driven intelligent system that can generalize to unseen scenarios.
Solution Approach 2:
The patent changes the fundamental parameters of the decision-making system by transitioning from fixed logic rules and cost functions to learned parameters from training data. The system learns optimal navigation policies by adjusting internal parameters through machine learning algorithms, enabling adaptation to diverse and tricky driving conditions that were previously unaccounted for.
2Adaptability or versatility
If machine learning models are trained on extensive vehicle sensor data to improve scenario coverage and reliability, then the system can handle tricky conditions better, but the device complexity and computational requirements increase
Solution Approach 1:
The patent applies preliminary action by training machine learning models offline using extensive vehicle sensor data collected during normal operation. The models are pre-trained on diverse scenarios including tricky conditions, and then deployed to autonomous driving systems. This allows the complex learning process to occur beforehand, enabling the onboard system to make rapid decisions without real-time training complexity.
Solution Approach 2:
The patent uses copying by training ML models on replicated and augmented versions of real-world sensor data. The system creates synthetic training scenarios that copy and extend actual driving conditions, allowing the model to learn from extensive varied data without requiring physical exposure to every possible tricky scenario, thereby reducing the complexity of data collection while maintaining high adaptability.
Data Source
AI summary
Systems and methods for generating safe driving paths in autonomous driving adversity conditions can include receiving, by a computer system including one or more processors, sensor data of a vehicle that is traveling. The sensor data can be indicative of one or more autonomous driving adversity conditions, and can include image data depicting surroundings of the vehicle. The method can include executing, by the computer system, a trained machine learning model to predict, using the sensor data, a trajectory to be followed by the vehicle through the one or more autonomous driving adversity conditions, and providing, by the computer system, an indication of the predicted trajectory to an autonomous driving system of the vehicle.