DASVIDANIA — AI for an autonomous acoustic drone

Jean-Daniel Cha­zot, a tenured pro­fes­sor in the Depart­ment of Mechan­i­cal Engi­neer­ing at the UTC’s Rober­val Lab­o­ra­to­ry and a con­tract lec­tur­er attached to UTC-Heudi­asyc togeth­er with Jean-Daniel Cha­zot, a mem­ber of the res­i­dent « Acoustics and Vibra­tions » team and Jos­sué Car­iño Esco­bar, a robot­ics engi­neer work­ing on aer­i­al drones, led the DASVIDANIA project, which was select­ed dur­ing the first call for pro­pos­als under the TWIN 2020 programme.

‘An ini­tial col­lab­o­ra­tion with Pedro Castil­lo of Heudi­asyc and Sor­bonne Uni­ver­si­ty, as part of a Cifre PhD fund­ed by CETIM, has brought the fields of acoustic imag­ing and drone con­trol clos­er togeth­er. This has result­ed in the devel­op­ment of an autonomous drone demon­stra­tor capa­ble of iter­a­tive­ly detect­ing acoustic sources, from the loud­est to the faintest,” explains Jean-Daniel Chazot.

A suc­cess­ful exper­i­ment, car­ried out by a sin­gle PhD stu­dent, which both teams wished to con­tin­ue. The call for pro­pos­als for TWIN 2026 pro­gramme there­fore came at just the right time, and DASVIDANIA (Acoustic Drone for Autonomous Sur­veil­lance and Detec­tion Pow­ered by Arti­fi­cial Intel­li­gence), as part of the CRI ‘Arti­fi­cial Intel­li­gence and Tech­no­log­i­cal Sys­tems’, was pro­posed to com­bine exper­tise in acoustic imag­ing with that in drone con­trol in order to improve the per­for­mance of the cur­rent drone.

Two PhD the­ses have been launched; one, super­vised by Alessan­dro Vic­tori­no and co-super­vised by Jos­sué Car­iño Esco­bar, focus­es on devel­op­ing a robust adap­tive con­trol sys­tem for a drone using an AI-based dis­tur­bance com­pen­sa­tion approach; the sec­ond, co-super­vised by Jean-Daniel Cha­zot and Gau­ti­er Lefeb­vre from UTC-Com­pieg­ne, as well as by François Ollivi­er from the Sor­bonne Uni­ver­si­ty, is ded­i­cat­ed to opti­mis­ing a drone-mount­ed micro­phone array aimed at track­ing an acoustic tar­get and to “noise can­celling” acoustic mea­sure­ments in the pres­ence of airflow.

“One of our objec­tives is to devel­op an autonomous drone equipped with an acoustic array and to inte­grate AI into it in a spe­cif­ic way. At UTC-Heudi­asyc, we had already begun using neur­al net­works for drone con­trol, but we want­ed to go fur­ther and inte­grate oth­er AI meth­ods, not only for the acoustic sen­sors but also for the con­trol sys­tem. The idea is to design a robust adap­tive con­troller fed in part by acoustic data. Thus, knowl­edge of wind direc­tion, veloc­i­ty or strength will enable us to improve con­trol effi­cien­cy and ensure greater sta­bil­i­ty for the drone,” says Jos­sué Car­iño Escobar.

What role does AI play in opti­mis­ing the acoustic anten­na? “From an acoustic per­spec­tive, we want to take things fur­ther. Cur­rent­ly, the demon­stra­tor is in an indoor flight cham­ber; the aim with DASVIDANIA is to devel­op a drone capa­ble of fly­ing out­doors, where it will be sub­ject­ed, in par­tic­u­lar, to wind noise that will degrade its acoustic per­for­mance. Not only the wind, but also air tur­bu­lence caused by the drone itself will dis­rupt the noise we wish to mea­sure and poten­tial­ly drown out the sig­nal to be detect­ed. The ’The idea is to detect wind noise – its direc­tion and inten­si­ty – and then sub­stract this noise from the mea­sured sig­nal. Until now, we have relied on con­ven­tion­al noise-can­celling tech­niques. In this project, we have decid­ed to incor­po­rate AI trained using exper­i­men­tal data col­lect­ed in an ane­choic cham­ber. ‘This will enable us, in a sec­ond phase, to sub­stract wind noise from in-flight acoustic mea­sure­ments,’ explains Jean-Daniel Chazot.

This work is being car­ried out in col­lab­o­ra­tion with the Sor­bonne Uni­ver­si­ty. “They have exten­sive exper­tise in acoustic mea­sure­ment anten­nas. In par­tic­u­lar, they have devel­oped the anten­na sys­tem based on ultra-light­weight MEMS micro­phones that we will be using in this project,” adds Jean ‑Daniel Chazot.

DASVIDANIA is of great impor­tance for improv­ing the drone’s per­for­mance in terms of detect­ing, locat­ing and track­ing an acoustic source. What are some poten­tial appli­ca­tions? “There are many, and they can address both civil­ian and mil­i­tary chal­lenges. The acoustic drone could, for exam­ple, be used to detect a gas leak in indus­tri­al net­works or to locate peo­ple in dis­tress,” says Jos­sué Car­iño Escobar.

Jean-Daniel Cha­zot adds: “The drone could also tack­le the wor­ry­ing issue of intru­sive drones fly­ing over sen­si­tive sites such as air­ports or nuclear pow­er plants or even detect and elim­i­nate inva­sive alien insects such as the Asian hornet.”

MSD

Le magazine

linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram