Trajectory generation apparatus, multi-link system, and trajectory generation method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing trajectory generation methods for multi-link mechanisms, such as robot arms, face challenges in efficiently generating trajectories that avoid obstacles while ensuring optimality and completeness, as they often struggle with converging learning due to complex multidimensional state spaces and large motion pattern sets.
Innovation Solution
A trajectory generation apparatus that uses machine learning to estimate a hand position at a subsequent time based on obstacle position coordinates, subject joint state, and target joint state, and then searches for a non-interfering joint state using this hand position as a restriction, thereby simplifying the learning process and improving convergence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning is used to generate trajectories in the original joint state space, then the system can learn motion patterns, but the learning convergence becomes difficult due to complex multidimensional state spaces
Solution Approach 1:
The patent segments the trajectory generation problem into two distinct spaces: a simplified hand position space for machine learning and an original joint state space for final trajectory execution. The hand position estimator learns in a reduced-dimensional space (3-6 dimensions for hand position and orientation) while the search unit handles the full joint state space, dividing the complex learning task into manageable parts that converge faster
Solution Approach 2:
The patent introduces hand position as an intermediary variable between the simplified learning space and the full joint state space. The hand position estimator produces hand position predictions that serve as constraints for the search unit, which then finds corresponding joint states. This intermediary layer allows the system to learn from simplified observations while ensuring valid joint configurations
2Reliability
If the system searches for non-interfering joint states without restrictions, then it ensures completeness of obstacle avoidance, but the search time increases significantly
Solution Approach 1:
The patent performs preliminary estimation of hand position using machine learning before conducting the obstacle avoidance search. By pre-computing the expected hand position and using it as a constraint for the search unit, the system narrows down the search space in advance, ensuring that the search only explores feasible regions that are likely to lead to valid trajectories
Solution Approach 2:
The patent applies different processing qualities to different parts of the trajectory generation process. The hand position estimation uses a simplified model with lower computational requirements, while the obstacle avoidance search focuses locally on finding joint states that satisfy the hand position constraint and avoid obstacles, rather than exhaustively searching the entire joint space
Data Source
AI summary
In a trajectory generation apparatus, position coordinates of an obstacle existing in a motion space of a robot arm is acquired. A hand position at a second time, which is a time next to a first time, is estimated by using a learning result of machine learning, based on the position coordinates of the obstacle, a subject joint state of the robot arm at the first time, and a target joint state of the robot arm. A non-interfering joint state of the robot arm at which the obstacle does not interfere with the robot arm at the second time is searched for by using the hand position as a restriction.