Power-Efficient Implementation of Ternary Neural Networks in Edge Devices

Molina, Miguel; Mèndez, Javier; Morales, Diego P.; Castillo, Encarnación; Vallejo, Marisa López; Cuéllar, Manuel Pegalajar

doi:10.1109/jiot.2022.3172843

Cited by 3 publications

(2 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In its final version, power optimized AI networks running on FPGA should take over this task. In this scope, a framework to train, optimize and implement such algorithms has been developed [13]. Furthermore, a self-pose estimation system running on the NVIDIA Jetson Nano is under development.…”

Section: Communication Systems Facilitating Bvlos Drone Servicesmentioning

confidence: 99%

Tackling different aspects of drone services utilizing technologies from cross-sectional industries

Purucker,

Reis,

Larsen

et al. 2023

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

Enabling autonomous and Beyond Visual Line of Sight (BVLOS) operation of Unmanned Aerial Vehicles (UAVs) in the Very Low Level (VLL) airspace requires further advancement of technologies such as sensing the environment or secure and reliable communication. This paper addresses these challenges by presenting solutions developed within the project Airborne Data Collection on Resilient System Architectures (ADACORSA). Here, findings from cross-sectional areas such as the automotive industry are being further enhanced to fulfill the demands of aviation, in particular for use in the UAV domain. The developed technologies include an advanced Ethernet-based deterministic network for reliable onboard communication, a multi-sensor architecture for sensing the spatial environment as well as a multi-link communication gateway that provides reliable communication to the ground and a secure handover architecture.

show abstract

Section: Communication Systems Facilitating Bvlos Drone Servicesmentioning

confidence: 99%

Tackling different aspects of drone services utilizing technologies from cross-sectional industries

Purucker,

Reis,

Larsen

et al. 2023

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

show abstract

“…This approach is particularly beneficial when hardware parallelization of the algorithms is possible, as with most of the Artificial Neural Networks (ANNs) topologies. For this reason, the first efforts have been focused into the creation a framework to train, optimize, and implement ANNs on FPGAs, as presented in [16]. The goal of this framework is to provide highly certified and sustainable solution for resource and energy-constrained devices.…”

Section: Hardware Co-design Power Optimizationmentioning

confidence: 99%

Multisensor Avionics Architecture for BVLOS Drone Services

Giovagnola

Molina

Beneder

et al. 2023

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

This ADACORSA demonstrator focuses on the implementation of a fail-operational avionics architecture combining Commercial Off-The-Shelf (COTS) elements from the automotive, the aerospace and the artificial intelligence world. A collaborative sensor setup (Time-of-Flight camera and FMCW RADAR from Infineon Technologies, stereo camera, LiDAR, IMU and GPS) allows to test heterogeneous sensor fusion solutions. A Tricore Architecture on AURIX™ Microcontroller supports the execution of safety supervision tasks as well as data fusion. A powerful embedded computer platform (NVIDIA Jetson Nano) accelerates AI algorithms performance and data processing. Furthermore, an FPGA enables power optimization of Artificial Neural Networks. Finally, a Pixhawk open-source flight controller ensures stabilization during normal flight operation and provides computer vision software modules allowing further processing of the captured, filtered and optimized environmental data. This paper shows various hardware and software implementations highlighting their emerging application within BVLOS drone services.

show abstract