Learned Trajectory Generation With Feedback for Target-Reaching Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing trajectory generation systems may produce trajectories that do not reach the target position, leading to the inability to decide on a high-quality trajectory.
Innovation Solution
A trajectory generation system that uses a learned model generated by machine learning to generate trajectories by inputting environmental feature information, start state, and target state, while allowing for variations in a first parameter such as dropout rate, to prevent collisions and produce multiple trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a plurality of trajectory candidates are generated leading to a target area including the vicinity of the target position, then the number of trajectory options increases, but the reliability of reaching the exact target position decreases
Solution Approach 1:
The patent implements a feedback mechanism where the evaluation unit assesses each generated trajectory and provides feedback to the computing unit. Trajectories that do not satisfy predetermined conditions (such as reaching the exact target position) are fed back for regeneration with adjusted parameters, while satisfactory trajectories are selected. This closed-loop feedback ensures both diversity in trajectory options and reliability in reaching the target position.
2Adaptability or versatility
If multiple trajectory candidates are generated with different degrees of freedom, then the versatility of trajectory options increases, but the complexity of evaluating and selecting the optimal trajectory increases
Solution Approach 1:
The evaluation unit autonomously evaluates each generated trajectory against predetermined conditions without requiring external intervention. The system self-regulates by automatically regenerating trajectories that fail evaluation and selecting those that succeed, reducing the need for complex external evaluation mechanisms and simplifying the overall system architecture.
3Manufacturing precision
If trajectories are generated using a learned model with fixed parameters, then the manufacturing precision of trajectory generation is high, but the adaptability to different environmental conditions decreases
Solution Approach 1:
The patent transforms the static learned model into a dynamic system by allowing the computing unit to regenerate trajectories with adjusted parameters when environmental conditions change or when generated trajectories do not satisfy evaluation conditions. This dynamic parameter adjustment enables the system to adapt to different environmental conditions while maintaining high trajectory generation precision through the underlying learned model.
Data Source
AI summary
A trajectory generation system capable of deciding a high-quality trajectory is provided. The trajectory generation system includes an environmental feature information acquisition unit and a trajectory generation unit. The environmental feature information acquisition unit acquires environmental feature information that indicates features of an environment around a moving object. The trajectory generation unit inputs a start state and a target state of the moving object and environmental feature information into a learned model that is generated by machine learning in advance and is used to generate a trajectory along which the moving object can move in the environment, and generates, for each of a plurality of different values of a first parameter that is set in the learned model and can be changed, a trajectory using the learned model.