Live Demonstration: Real-time neuro-inspired sound source localization and tracking architecture applied to a robotic platform

Li e Demons a ion: Real- ime neu o-inspi ed sound sou ce localiza ion and acking a chi ec u e applied o a obo ic pla o m F. Pe ez-Peña1 E. Ce ezuela-Escude o2, Angel Jimenez-Fe nandez2 and A u o Mo gado-Es e ez1 1. Applied Robo ics Resea ch Lab, Uni e sidad de Cádiz, Facul y o Enginee ing, Pue o Real, Cadiz, Spain 2. Robo ics and Technology o Compu e s Lab (RTC), Uni e si y o Se ille, ETSI In o má ica, Se ille, Spain Email: [email p o ec ed] Abs ac — This li e demons a ion p esen s a sound sou ce localiza ion and acking sys em implemen ed wi h Spike Signal P ocessing (SSP) building blocks on FPGA de ices. The sys em a chi ec u e is based on he abili y o he mammalian audi o y sys em o loca e he di ec ion o a sound in he ho izon al plane using he in e au al in ensi y di e ence. We used a binau al Neu omo phic Audi o y Senso o ob ain spike a es simila o hose gene a ed by he inne hai cells o he human audi o y sys em and he componen ha ob ains he in e au al in ensi y di e ence is inspi ed by he la e al supe io oli e. The spike s eam ha ep esen s he in e au al in ensi y di e ence is used o u n a obo ic pla o m owa ds he sound sou ce di ec ion. The sys em was es ed wi h pu e ones (1-kHz, 2.5-kHz and 5- kHz sounds) wi h an a e age e o o 2.32 deg ees. I. I NTRODUCTION This li e demo is based on e e ence [1]. I shows he ha dwa e implemen a ion o a sound localiza ion and acking sys em inspi ed by he mammalian audi o y sys em. The NAS senso used p oduces a biological cochlea-like ou pu . This ou pu is he s imulus o he p ocessing sys em whe e he LSO model is implemen ed. The a chi ec u e p oposed o he LSO which pe o ms he sub ac ion be ween wo inpu spike a es p oduces he In e au al In ensi y Di e ence (IID). The IID audi o y cue is used as he inpu o he spike-based ac ua ion s age ha acks he sound. The demo shows he sys em es ed using 1 kHz, 2.5 kHz and 5 kHz pu e ones. The maximum e o ob ained is less han i e deg ees. Fu he mo e, ou sys em shows a high noise ole ance le el when whi e noise is applied: in he wo s condi ion, he a e age e o is lowe han en deg ees. The a chi ec u e p esen ed in his demo is implemen ed by using low-cos comme cial ha dwa e de ices such as FPGAs. The powe consump ion goes up o 58.33 mW in ope a ion (29.7 mW om he NAS and 28.63 mW om he p ocessing laye ). II. D EMOSTRATION SETUP The expe imen al se up is shown in Fig. 1. I consis s o a s imulus (sound sou ce), a obo ic pla o m (head), an audi o y senso and an ac ua ion laye . The dis ance be ween he speake and he head is 40 cm a di e en azimu hal angles (0º o 90º in s eps o 15º). The mic ophones a e on each side o he head (omnidi ec ional pick-up pa e n). The head is placed on op o a pla o m d i en by a DC mo o wi h an encode (Mic omo o Re . 2224R006SR plus gea head Re . 20/1 112:1 and encode Re . IE2-512 om Faulhabe ). The NAS is implemen ed using a Vi ex5 FPGA (XC5VFX70T) and i uses up o 99% o he o al slices a ailable. The FPGA is in a Xilinx de elopmen boa d (ML507) which includes he AC’97 audio codec. The NAS ou pu is connec ed o he p ocessing sys em using he AER p o ocol. The p ocessing sys em is also implemen ed using a Vi ex5 FPGA (XC5VFX30T), which uses up o 4% o he o al slices a ailable. The speci ica ions o he mic ophone a e: ansduce p inciple based on back elec e condense elemen , he equency esponse ange is be ween 20 and 16,000Hz, he sensi i e is -64dB ±3dB and he impedance is 1,000 Ohm. Fig 1. Se -up p o ided o he isi o expe ience. III. V ISITOR EXPERIENCE Visi o s will be able o in e ac wi h he demo by mo ing he audio sou ce wi hin he ange (-90, 90) deg ees and by modi ying he one played by he sou ce: (1, 2.5 and 5) kHz. They will check i he head is able o ollow he sou ce and how accu a e is he mo emen . The e will be a small sc een o show he g ound u h and a lap op showing he cu en posi ion eached by he head (Figu e 1). A CKNOWLEDGMENT This wo k is suppo ed by he Spanish g an (wi h suppo om he Eu opean Regional De elopmen Fund) COFNET (TEC2016-77785-P). IV. R EFERENCES [1] E. Ce ezuela-Escude o, F. Pé ez-Peña e al., “Real- ime neu o-inspi ed sound sou ce localiza ion and acking a chi ec u e applied o a obo ic pla o m,” Neu ocompu ing, ol. 283C, pp. 129-139, Feb. 2018.