Ab­stract

Cor­rect and com­plete per­cep­tion is key for au­tonomous ve­hi­cles to plan safe ma­neu­vers. Es­pe­cially under harsh weather con­di­tions the use of sens­ing ca­pa­bil­i­ties from other road users via ve­hi­cle-to-every­thing (V2X) com­mu­ni­ca­tion can con­tribute to more com­plete per­cep­tion. How­ever, in­for­ma­tion from other road users may con­tain ad­di­tional un­cer­tain­ties and lead to less ac­cu­rate per­cep­tion. Ad­di­tion­ally, at­tack­ers may use the V2X chan­nel to trans­mit ma­li­cious data. For build­ing an ac­cu­rate en­vi­ron­men­tal model an au­tonomous ve­hi­cle needs as pre­cise in­for­ma­tion as pos­si­ble. To tackle the prob­lems of ad­di­tional un­cer­tain­ties within col­lec­tive per­cep­tion we pro­pose a method­ol­ogy to check per­ceived in­for­ma­tion for its trust­wor­thi­ness and va­lid­ity. This is achieved by eval­u­at­ing the per­cep­tion ca­pa­bil­i­ties of a holis­tic per­cep­tion pipeline and check­ing col­lec­tively trans­mit­ted in­for­ma­tion for con­sis­tency. The pro­posed ap­proach is eval­u­ated under vary­ing en­vi­ron­men­tal con­di­tions on a sim­u­lated high­way sce­nario.