Neural networks improve real-time quadrotor trajectory accuracy
A Neural Network Approach towards Real-Time Knot Point Density Estimates for the Nonlinear Quadrotor
Machine Learning
Summary
Quadrotors follow paths that can be tricky to describe efficiently with simple math tools. The authors found that placing points used to build these path descriptions evenly doesn't work well when the path changes quickly. They used a neural network to predict where to put more points for better accuracy. This approach lets the quadrotor's path be described more precisely using fewer points in real time.
What this means in practice
- •For drone control engineers: Generate efficient flight paths with fewer control points for nonlinear quadrotors by predicting adaptive knot placements in real time.
- •For robotics simulation developers: Improve trajectory approximation accuracy in simulators by using learned adaptive knot placement for dynamic systems.
Authors
Jasmine Ratchford, Raffaele Romagnoli, Mark H. Klein
Abstract
Spline representations provide a compact way to approximate continuous trajectories using a finite number of parameters. When approximating trajectories generated by dynamical systems, the efficiency of a spline representation depends strongly on the placement of knot points. Uniform knot distributions often lead to inefficient representations, requiring either unnecessary control points or producing poor approximations in regions where the trajectory varies rapidly. In this work we propose a learning-based method for adaptive knot placement in B-spline approximations of trajectories generated by ordinary differential equations. A neural network learns a mapping from the initial condition of the system to quantities that characterize the local variation of the resulting trajectory. These estimates are used to construct nonuniform knot distributions that approximately equidistribute a local error indicator, reducing the number of control points needed to meet a prescribed accuracy.