ACROS: the robots of the future need a brain. Mag­a­zino devel­ops it. 

2. October 2018


ACROS stands for “Advanced Coop­er­a­tive Robot Oper­at­ing Sys­tem” and is the oper­at­ing sys­tem we devel­oped for Magazino’s per­cep­tion-con­trolled robots. Our robots are also deployed, coor­di­nated and con­trolled through ACROS. Addi­tion­ally, ACROS can assist in the devel­op­ment of other per­cep­tion dri­ven robots with dif­fer­ent capa­bil­i­ties. Thus it has the poten­tial to gen­er­ate a global knowl­edge data­base from the data of all robots oper­ated with ACROS: a basis for machine learn­ing and thus new intel­li­gent robot behaviour.

In the past, robots were very deter­min­is­tic – you could also say: hard-work­ing, but stu­pid. Once pro­grammed, they put a weld­ing spot in the same place thou­sands of times at a fixed rate. More­over, since these robots had no per­cep­tion of their sur­round­ings, let alone the abil­ity to react intel­li­gently to them, they had to work sep­a­rated from humans by fences and light bar­ri­ers. This has changed fun­da­men­tally. Through the use of numer­ous sen­sors, 3D cam­eras and com­puter vision, next-gen­er­a­tion robots are now bet­ter able to under­stand what is hap­pen­ing in their envi­ron­ment. At the same time, the envi­ron­ment for robots is becom­ing increas­ingly com­pli­cated, espe­cially if they have to work with peo­ple in the same area. The two impli­ca­tions of this are that the envi­ron­ment can change rapidly, and that the robot must act safely at all times. Exist­ing pro­gram­ming tools and solu­tions, such as a PLC con­trol sys­tems, can no longer meet these require­ments. We need rad­i­cally new approaches to pro­gram­ming, con­trol­ling and net­work­ing robots. This is because next-gen­er­a­tion robots are no longer deter­min­is­tic, but per­cep­tion-con­trolled. That’s why we devel­oped ACROS. One could say that ACROS is some­thing like a brain for robots. Our robots TORU and SOTO are the first real-world instances of robots empow­ered by ACROS.

»We need radically new ways to code, control and connect robots.«
ACROS means: new possibilities

In the past, robots were always pro­grammed by their indi­vid­ual man­u­fac­tur­ers with cus­tom soft­ware,  exclu­sively tai­lored to their respec­tive needs.
ACROS can do more. ACROS makes it pos­si­ble for the first time to pro­gram dif­fer­ent types of robots with a com­pre­hen­sive “oper­at­ing sys­tem”. This would allow the phys­i­cal (hard­ware) com­po­nents of robots to become inter­change­able in the future — the intel­li­gence lies in the ACROS soft­ware frame­work. And this is where there is extra­or­di­nary poten­tial. As more robots world­wide work with ACROS, more knowl­edge is col­lected in the cloud. Through this net­work, the robots can learn from one another and con­stantly improve. ACROS also grows with col­lec­tive intel­li­gence. Our plan is to develop ACROS into a global plat­form where devel­op­ers can add their own appli­ca­tions and interfaces.

And what might that look like in the future?

ACROS makes per­cep­tion dri­ven robot­ics pos­si­ble. If ACROS is devel­oped suf­fi­ciently, ACROS cus­tomers could ben­e­fit from the pos­si­bil­ity of using a pre-exist­ing frame­work for their  devel­op­ment. This saves resources and accel­er­ates the process enormously.
ACROS will be com­pat­i­ble with numer­ous types of robots. Only hard­ware-spe­cific com­po­nents of ACROS need to be adapted, such as for con­trol­ling a robotic manip­u­la­tor. Another major ben­e­fit will be that all robots work­ing with ACROS will be able to share their “expe­ri­ence” in the cloud. You can imag­ine it like this: Using the data that Robot A col­lects when pick­ing shoe boxes, Robot B can also pick arbi­trary objects that are phys­i­cally sim­i­lar, with­out ever hav­ing encoun­tered such objects before.

How does ACROS work in practice?

A (still) fic­ti­tious prac­ti­cal exam­ple: A cus­tomer from the auto­mo­tive indus­try would like a soft­ware solu­tion with which totes can be located on a con­veyor belt with com­puter vision and then gripped and stacked with a robot arm. Many mod­ules, such as sta­tis­tics and error report­ing, which were pro­grammed for TORU could be used sim­ply by drag-and-drop to pro­gram the exter­nal hard­ware. ACROS can also be a very eco­nom­i­cal mod­u­lar sys­tem for pro­gram­ming. How­ever, the ben­e­fits and poten­tial of ACROS go far beyond this. ACROS can be used for remote main­te­nance, fleet man­age­ment, as a data­base for objects and orders, and as a com­mand cen­tre in a warehouse.

ACROS: in a nutshell

Advanced Coop­er­a­tive Robots are per­cep­tion-con­trolled: ACROS must process cam­era and sen­sor data in real time and trig­ger cor­re­spond­ing reactions.

Advanced Coop­er­a­tive Robots have to deal with uncer­tainty and work smoothly in par­al­lel: ACROS bun­dles the data of the entire fleet in oper­a­tion and imme­di­ately com­mu­ni­cates envi­ron­men­tal changes to all robots.

Advanced Coop­er­a­tive Robots can learn on their own: the ACROS net­work and the con­nected global data­base enable machine learn­ing and thus intel­li­gent behaviour.

The core of ACROS is the unique con­cept of mod­u­lar seman­tic mod­els. This allows its use across dif­fer­ent robotic platforms:

  • The mod­u­lar archi­tec­ture of ACROS is a com­pre­hen­sive toolkit for pro­gram­ming per­cep­tion-con­trolled and AI-enabled robotic applications.
  • The soft­ware com­po­nents range from hard­ware dri­vers, per­cep­tion and motion algo­rithms to task ful­fill­ment and fleet coor­di­na­tion. The cloud infra­struc­ture for global robot knowl­edge exchange and machine learn­ing is also included.
  • The cen­tral con­cept here is the sep­a­ra­tion of code from seman­tic mod­els of the envi­ron­ment, the robot, the objects to be processed and the tasks to be performed.
  • This approach allows easy cus­tomiza­tion for dif­fer­ent robotic plat­forms, envi­ron­ments, and tasks, enabling rapid deploy­ment and seam­less operation.