Robot Trajectory Selection Using Weighted Similarity Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for selecting potential trajectories in autonomous vehicles rely on random selection or ensemble models, which can lead to inefficient and unsafe planning due to unnecessary diversity or lack of diversity in future scenarios.
Innovation Solution
A method is introduced to determine potential trajectories by optimizing a metric using machine learning models, determining weighted similarity values, and adapting the set of proposed trajectories to minimize error, ensuring robust and safe planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If random selection or ensemble models are used to select potential trajectories, then a predefined number of trajectories can be obtained, but the planning efficiency is reduced and safety is compromised due to unnecessary diversity or lack of diversity in future scenarios
Solution Approach 1:
The patent changes the selection criterion from random selection or simple ensemble voting to optimization based on a metric that balances diversity and relevance. The selector optimizes a metric that considers both the diversity of trajectories and their relevance to the current situation, thereby improving planning efficiency while maintaining safety through scientifically grounded selection rather than random or purely democratic approaches.
2Reliability
If all potential trajectories are considered in planning, then comprehensive safety can be achieved, but the planning time increases significantly
Solution Approach 1:
The patent extracts only the most relevant trajectories from the full set of potential trajectories by using a selector that optimizes a metric balancing diversity and relevance. This extraction process identifies and focuses on the critical subset of trajectories that matter for safe planning, discarding unnecessary ones, thereby reducing planning time while maintaining comprehensive safety coverage for the important cases.
3Reliability
If diverse potential trajectories are selected, then robust planning can be ensured, but unnecessary diversity is introduced in unimodal future scenarios
Solution Approach 1:
The patent implements a dynamic selection process where the metric optimized by the selector adapts to the characteristics of the current situation. In diverse future scenarios, the metric allows for greater trajectory diversity to ensure robustness, while in unimodal scenarios, it automatically reduces diversity to avoid unnecessary complexity. This dynamic adaptation ensures robust planning only when and where it is actually needed.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
The invention relates to a method (200) for controlling a robot device, comprising: determining a plurality of potential trajectories of an object in the environment of the robot device by using one or more machine learning models to determine one or more potential trajectories and, for each of these, an associated weighting factor; determining a plurality of weighted similarity values by determining a weighted similarity value for each of the plurality of potential trajectories, the determination comprising: determining a similarity value which represents a similarity between the potential trajectory and a set of proposed trajectories according to at least one similarity metric, and determining the weighted similarity value by weighting the similarity value according to the weighting factor associated with the potential trajectory;Adapting the set of proposed trajectories to determine an adapted set of proposed trajectories that results in a reduced sum of the multitude of weighted similarity values; generating control parameters for controlling the robot device using the adapted set of proposed trajectories; and controlling the robot device according to the control parameters.