Trajectory Initialization for Autonomous Vehicle Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous vehicle planning systems often over- or under-estimate the vehicle's state, leading to abrupt acceleration or deceleration, failure to stop completely, or other control issues that result in an unpleasant ride, unnecessary wear, and potential non-compliance with traffic rules.
Innovation Solution
The system employs a multi-layer planning approach with a decision planning system and a drive planning system operating at different frequencies to generate and refine vehicle trajectories, using sensor data to determine better initialization states and smooth out control inputs, ensuring a comfortable and safe ride.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional models are used to estimate vehicle state, then the planning system can generate control trajectories, but the model may over- or under-estimate the actual state leading to abrupt acceleration or deceleration
Solution Approach 1:
The system implements feedback by continuously comparing the actual vehicle state (from sensors) with the estimated state (from the model) and using this difference to correct trajectory initialization. The planning system receives feedback about estimation errors and adjusts subsequent trajectory generation to compensate for model inaccuracies, preventing abrupt control actions.
Solution Approach 2:
The system performs preliminary action by pre-initializing trajectories using the model's estimated state before executing control commands. This allows the system to prepare smooth transition paths in advance based on predicted vehicle state, ensuring that control inputs are gradual and comfortable rather than abrupt when the actual state is reached.
2Productivity
If conventional trajectory generation is used, then control trajectories can be produced, but abrupt acceleration or deceleration occurs resulting in unnecessary wear on the vehicle
Solution Approach 1:
The system applies dynamics by making the trajectory initialization adaptive rather than static. The initialization state is dynamically adjusted based on the difference between actual and estimated vehicle states, allowing the trajectory generation to smoothly adapt to real conditions. This dynamic adjustment prevents abrupt control commands that would cause vehicle wear while maintaining efficient trajectory generation.
3Adaptability or versatility
If conventional models estimate vehicle state, then trajectories can be generated, but the vehicle may fail to comply with traffic rules and regulations
Solution Approach 1:
The system uses feedback from actual vehicle state measurements to ensure reliable traffic rule compliance. By continuously monitoring the difference between estimated and actual states, the system can detect when model predictions would lead to non-compliance and adjust trajectories accordingly, ensuring that control actions always meet traffic requirements regardless of model estimation errors.
Data Source
AI summary
Trajectory determination for controlling a vehicle, such as an autonomous vehicle, is described. In an example, a vehicle system includes multiple planning systems for calculating trajectories. A first system may calculate first trajectories at a first frequency and the second system may calculate second trajectories at a second frequency and based on the first trajectories. The first and/or second trajectories may be initialized at states of the vehicle corresponding to a projection onto a previous-in-time respective first or second trajectory. The second trajectories may be control trajectories along which the vehicle is controlled.