Core Concepts
FPGAs offer a solution to computational demands for UAVs, enhancing adaptability and performance.
Abstract
Introduction to UAVs and their applications in various fields.
Importance of onboard computing advancements for UAVs.
Utilizing DNNs with RL to enhance UAV performance.
Challenges of implementing DNNs on UAVs due to limited computing power.
Exploring FPGAs as a solution for computational demands.
Implementation of DNNs on an FPGA-equipped expansion deck.
Detailed explanation of the LF expansion deck design and functionality.
Utilizing hardware accelerators for faster onboard computation.
Implementation of fixed-point arithmetic for DNN weight conversion.
Discussion on the dynamic model of the quadrotor and the DNN model.
Validation of the implementation through flight trajectory experiments.
Conclusions on successful DNN implementation and future work.
Stats
"The advent of unmanned aerial vehicles (UAVs) has improved a variety of fields by providing a versatile, cost-effective and accessible platform for implementing state-of-the-art algorithms."
"We propose a novel hardware board equipped with an Artix-7 FPGA for a popular open-source micro-UAV platform."
"The quadrotor is modeled as a rigid body, where x denotes the position of its center of mass in an inertial reference frame."
"The self observation vector is defined as oq t = (pq,j t , vq t, rq t, ωq t)."
"The output of the neural network is defined as a ∈[−1, 1], and we obtain the thrust vector by applying an affine transformation ˆ f = 1 2(clip(a, −1, 1)+1)."
Quotes
"FPGAs offer a promising solution to the computational demands of deploying DNNs on UAVs."
"The trained neural network is capable of low-level motor control while avoiding collisions among the quadrotors."