.png){kind=icon}
.png){kind=icon}
Return
Collision Avoidance for a Hexacopter Using Deep Learning and IMU in a GPS-Denied Environment
.png){kind=icon}
PRE
AI
K
L
J
DOI：10.1007/s12555-026-00052-3.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper proposes a deep neural network to improve the state estimation and collision avoidance performance of a hexacopter Unmanned Aerial Vehicle operating in GPS-denied environments. The proposed Pseudo-Measurement Network (PM-Net) takes the current hexacopter state, control thrust, and Inertial Measurement Unit (IMU) measurements as input and predicts the bias of the IMU acceleration measurement at each time step. The predicted bias is utilized to correct the acceleration measurement, which mitigates the drift in the state estimation of position and velocity. For collision avoidance, a collision cone-based algorithm with a Proportional-Derivative controller is proposed, using obstacle information obtained from an onboard RGB-D camera. The complete framework, including controller, PM-Net-based state estimation, and collision avoidance, is evaluated through numerical simulations, including Monte Carlo simulations. The results show that PM-Net improves the accuracy of state estimation compared with conventional IMU dead-reckoning, demonstrating its effectiveness for hexacopter operation in GPS-denied environments.
Keywords:
Collision avoidance
Deep learning
Inertial measurement unit
Unmanned aerial vehicle
Journal
IF:
2.9
Papers:
170
Citations:
6.5K
Organization
Cited Papers