Vehicle Trajectory Generation With GAN-Based Human-Like Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles (AVs) generate technically correct but unnatural robotic trajectories, leading to an unpleasant riding experience for passengers due to aggressive acceleration processes, which differ from human-driven vehicle trajectories.
Innovation Solution
A machine-learning (ML) based trajectory generator, trained using a generative adversarial network (GAN), refines heuristic-based trajectories to generate human-like trajectories that mimic the driving styles of human drivers, ensuring a more comfortable ride while maintaining safety by performing a pre-determined safety check.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If heuristic-based trajectories are used for autonomous vehicle navigation, then safety and technical correctness are ensured, but the riding experience becomes unnatural and unpleasant due to aggressive acceleration processes
Solution Approach 1:
The patent introduces an intermediary system (the trajectory generation system with GAN-based refinement) that mediates between the safe but robotic heuristic trajectories and the desired natural human-like driving behavior. The system refines the heuristic trajectories through machine learning models trained on human driving data, producing trajectories that maintain safety constraints while achieving natural acceleration profiles and riding comfort.
2Ease of operation
If machine learning refinement is applied to generate human-like trajectories, then riding comfort and naturalness are improved, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training the GAN-based machine learning models offline using extensive human driving data. The models are prepared in advance to recognize and refine trajectory patterns, so that during actual vehicle operation, the refinement process can proceed efficiently without adding real-time computational complexity to the control system.
3Adaptability or versatility
If GAN-based machine learning models are used to refine trajectories, then human-like driving characteristics are achieved, but computational resources and training time are increased
Solution Approach 1:
The GAN-based models are trained offline in advance using large datasets of human driving behavior. This preliminary training allows the system to learn complex human-like driving characteristics without consuming real-time computational resources during vehicle operation, thereby reducing the effective training time impact on operational performance.
4Reliability
If heuristic-based trajectories are used as fallback, then safety is maintained, but the naturalness of human-like trajectories is compromised
Solution Approach 1:
The patent implements a dynamic trajectory selection mechanism that adapts between different trajectory generation modes. The system dynamically switches between using refined human-like trajectories for normal operation (prioritizing naturalness) and falling back to heuristic-based trajectories when safety constraints require (prioritizing reliability). This dynamic adaptation allows the system to optimize for naturalness whenever possible while maintaining safety as the ultimate constraint.
Data Source
AI summary
In one embodiment, a computing system of a vehicle may access sensor data associated with a surrounding environment of a vehicle. The system may generate, based on the sensor data, a first trajectory having one or more first driving characteristics for navigating the vehicle in the surrounding environment. The system may generate a second trajectory having one or more second driving characteristics by modifying the one or more first driving characteristics of the first trajectory. The modifying may use adjustment parameters based on one or more human-driving characteristics of observed human-driven trajectories such that the one or more second driving characteristics satisfy a similarity threshold relative to the one or more human-driving characteristics. The system may determine, based on the second trajectory, vehicle operations to navigate the vehicle in the surrounding environment.