HC2022.ethz.AlfioDiMauro.v2.pdf

1、pulp-platform.orgpulp_ PlatformOpen Source Hardware,the way it should be!Kraken:A Direct Event/Frame-Based Multi-sensor Fusion SoC for Ultra-Efficient Visual Processing in Nano-UAVsAlfio Di Mauro(adimauroethz.ch)Moritz Scherer(scheremoethz.ch)Davide Rossi(davide.rossiunibo.it)Luca Benini(lbeniniethz

2、.ch)Toward nano and pico-size form factor UAVs1 A.Bachrach,“Skydio autonomy engine:Enabling the next generation of autonomous flight,”IEEE Hot Chips 33 Symposium(HCS),2021https:/www.bitcraze.io/products/crazyflie-2-13D Mapping&Motion Planning Object recognition&Avoidance0.06m2&800g of weight800g of

3、weightEnergy Capacity(Battery)5410mAh5410mAhApplicationsSearch&rescuePost-disaster inspectionSurveillancemaintenanceSmaller form factor of 0.008m2Weight of 27g 27g(30X lighter)Battery capacity of 250mAh250mAh(20X smaller)Advanced autonomous droneAdvanced autonomous droneNanoNano-dronedroneCan we fit

4、 sufficient“intelligence”in a 30X30X smaller payload and 20X20X lower energy budget?https:/ in tight space constraints?208/22/2022-Alfio Di Mauro-adimauroethz.chAchieving true autonomy on nano-UAVs Execute complex visual task at high speed and robustness fully on boardObstacle avoidance&NavigationOb

5、stacle avoidance&NavigationEnvironment explorationEnvironment explorationObject detectionObject detection308/22/2022-Alfio Di Mauro-adimauroethz.chAutonomous navigation building blocks deployable on KrakenRISCRISC-V FC:V FC:RGB frames from CPI sent to CUTIE and the RGB frames from CPI sent to CUTIE

6、and the RISCRISC-V ClusterV ClusterEventEvent-Frames from DVSI streamed to SNEFrames from DVSI streamed to SNERISCRISC-V Cluster:V Cluster:“DroNetDroNet”Obstacle avoidance network 2”Obstacle avoidance network 2 SNE:SNE:“LIF“LIF-FireNetFireNet”Low”Low-Latency Optical flow spiking Latency Optical flow