Autonomous Trajectory Scoring in Mixed-Reality Driver Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training and testing autonomous systems in real-world environments is unsafe due to the potential for accidents caused by untrained virtual drivers.
Innovation Solution
A simulator is used to train and test virtual drivers in a simulated environment, generating realistic scenarios and evaluating performance using a mixed-reality, closed-loop system to improve trajectory selection and safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training and testing are conducted in real-world environments, then the virtual driver learns from actual conditions, but safety risks increase due to potential accidents
Solution Approach 1:
The patent creates a simulated environment that copies real-world driving conditions, allowing the virtual driver to be trained and tested on actual road layouts, weather conditions, and traffic patterns without the safety risks of real-world deployment. The simulation replicates sensor inputs, environmental conditions, and dynamic objects to provide realistic training data.
2Measurement precision
If diverse and safety-critical scenarios are simulated, then the virtual driver's decision-making accuracy improves, but the complexity of the simulation system increases
Solution Approach 1:
The system pre-generates a diverse library of training scenarios including safety-critical situations before training begins. These scenarios are prepared in advance with all necessary sensor data, environmental conditions, and ground truth labels, allowing the virtual driver to be trained on comprehensive edge cases without requiring complex real-time scenario generation during training.
3Productivity
If automatic updates are implemented to improve virtual driver performance, then training efficiency increases, but the need for extensive real-world data collection and processing increases
Solution Approach 1:
The system implements automatic updates where the virtual driver is continuously retrained on newly generated or added simulation scenarios without requiring manual intervention. The training pipeline automatically processes new scenario data, updates the model weights, and validates performance, enabling self-improving capability that eliminates the need for manual data collection and processing cycles.
Data Source
AI summary
Trajectory value learning for autonomous systems includes generating an environment image from sensor input and processing the environment image through an image neural network to obtain a feature map. Trajectory value learning further includes sampling possible trajectories to obtain a candidate trajectory for an autonomous system, extracting, from the feature map, feature vectors corresponding to the candidate trajectory, combining the feature vectors into the input vector, and processing, by a score neural network model, the input vector to obtain a projected score for the candidate trajectory. Trajectory value learning further includes selecting, from the candidate trajectories, the candidate trajectory as a selected trajectory based on the projected score, and implementing the selected trajectory.