Robot Trajectory Generation Using Neural Cost-to-Go Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional motion planning for robots is computationally burdensome due to extensive collision checking and iterative propagation of cost values, making it inefficient for practical applications.
Innovation Solution
A method using higher-order function neural networks to generate a continuous set of values representing distances between configurations, allowing for faster trajectory generation by applying these values to a cost-to-go function, thereby partitioning the configuration space into collision and collision-free regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional motion planning algorithms are used to generate robot trajectories, then collision avoidance and configuration space partitioning are achieved, but computational burden increases significantly due to extensive collision checking and iterative propagation
Solution Approach 1:
The patent replaces traditional mechanical motion planning algorithms with a neural network-based system. The neural network is trained to predict collision-free trajectories and configuration space partitions, substituting iterative computational mechanics with a trained predictive model that provides comparable reliability with significantly reduced computational burden during execution.
Solution Approach 2:
The patent performs preliminary training of the neural network offline using extensive collision checking and configuration space analysis. This preliminary action pre-computes the knowledge needed for collision avoidance, so that during actual robot operation, the pre-trained network can quickly predict safe trajectories without performing extensive real-time collision checking.
2Reliability
If extensive collision checking and iterative propagation are performed to ensure accurate trajectory generation, then collision-free paths are guaranteed, but computation time increases
Solution Approach 1:
The patent substitutes iterative propagation algorithms with a neural network that directly predicts trajectory values and configuration space partitions. The network was trained to learn accurate collision-free paths, allowing it to provide reliable trajectory predictions without performing time-consuming iterative propagation during execution.
Solution Approach 2:
The patent changes the computational parameters from iterative numerical propagation to direct neural network inference. By transforming the problem into a pattern recognition task that the neural network can solve through learned parameters rather than iterative computation, the system maintains trajectory accuracy while dramatically reducing computation time.
Data Source
AI summary
A method for generating a trajectory of a robot from a first configuration to a second configuration within an environment while steering away from obstacles may include obtaining physical workspace information associated with the environment in which the robot is configured to operate; obtaining, using a first neural network, a set of weights of a second neural network that is configured to generate a set of values associated with a set of configurations of the robot with respect to the second configuration; obtaining, by applying the set of weights to the second neural network, the set of values associated with the set of configurations of the robot with respect to the second configuration; and generating the trajectory of the robot from the first configuration to the second configuration within the environment, based on the set of values.