Mag­a­zino awarded 1st place in “The Spark – the Ger­man Dig­i­tal Prize”

With ACROS, robots like TORU are able to fully autonomously iden­tify, grab and deliver items on the shelves ordered online by e‑commerce mer­chants in unit load ware­houses. Thanks to numer­ous sen­sors and 3D cam­era data, they are also able to work along­side the employ­ees on site. TORU is used, for exam­ple, in shoe ware­houses of logis­tics ser­vice providers such as Fiege Logis­tik or at the fash­ion mail-order com­pany Zalando, where it sup­ports employ­ees in the stor­age and retrieval of shoes ordered online.

ACROS sup­ports the devel­op­ment of intel­li­gent and mobile robots as well as their imple­men­ta­tion in the field with its mod­u­lar struc­ture. ACROS also serves to coor­di­nate and con­trol entire robot fleets. A global knowl­edge data­base, which gen­er­ates knowl­edge from data of all ACROS oper­ated robots (Deep Learn­ing), pro­vides the basis for intel­li­gent robot behav­iour (A.I.).

Until now, robots have always been indi­vid­u­ally pro­grammed by their man­u­fac­tur­ers accord­ing to the respec­tive require­ments with their own soft­ware. ACROS makes it pos­si­ble for the first time to pro­gram very dif­fer­ent types of robots quickly and eas­ily with a com­pre­hen­sive “oper­at­ing sys­tem”. Robots as hard­ware are thus inter­change­able – the intel­li­gence lies rather in the code. ACROS is there­fore less a sin­gle piece of soft­ware than a com­plete soft­ware stack. This includes, for exam­ple, the pro­gram mod­ules for object recog­ni­tion, free nav­i­ga­tion in the ware­house, plan­ning of grip­per move­ment and com­mu­ni­ca­tion of work orders with the inven­tory man­age­ment sys­tem in the ware­house. ACROS goes far beyond being a mod­u­lar sys­tem for pro­gram­ming. It is also used for remote main­te­nance, fleet man­age­ment, as a data­base for objects and orders and as a com­mand cen­tre in the warehouse.

What’s more, the more robots that work with ACROS, the more knowl­edge and data will reach the cloud. Through deep learn­ing, the robots con­stantly learn from each other, regard­less of where they are deployed or the objects they grip. The per­for­mance of the robots thus increases not only due to shorter times for object iden­ti­fi­ca­tion and local­iza­tion, but also because the robots can han­dle com­pletely unknown or com­plex sit­u­a­tions more reli­ably and confidently.