South American military laboratory developing HRED «brains» to combat bots, which are able to move confidently and make voice commands in the chaos of combat.
Bots are playing a huge role on the modern battlefield. Exploration, perimeter protection, the fight against improvised explosive devices, liquidation defense, armor and much more — all unmanned systems do even better manned. But modern boats require constant operator control, and if in the UAV there is progress towards automatic machines, then in ground combat bots tangible progress is slower. Grunt once «play» with joysticks and look on the screen — it’s just not safe in the criteria deadly close combat.
Experts behold the beginning of the development of unmanned systems with a simple set of «soldier’s mind», in other words capable of without the help of others to move over rough terrain, use tactics, talk with other robots, take orders given voice. This is enough for the more conventional method of benign and commander-human interaction and the slave machine.
But to make such an artificial intelligence at the technical level is very difficult. Specialists HRED together with the Municipal Institute Towson since 2004 working on a system SS-RICS, which connects voedinyzhdy sismvolicheskie subsimvolicheskie and knowledge into a single management structure. This system is applied based on the nature and cognitive architecture ACT-R, developed at the Institute of Michigan.
Specialists HRED long time trying to adapt to the military use of the latest generation in computer technology, simulating the natural processes of the human brain. As a result, unable to find out what exactly is necessary in order to achieve adequate bot of autonomous behavior in a variety of combat criteria.
Scientists have concluded that you must first reproduce some nuances zaniya first long-term memory and perception, then you can work on the behavior of generalized mental combat vehicle.
The knowledge comes as a result of different algorithms with different functions. Which specifically, it is easy to understand if for a moment to think about how we are trying to discover the latest product packaging. Memory reproduces images similar types of packaging, hands groping box, eyes seek out the band separation, etc. Brain produces complex overarching work, consisting of a large number of individual algorithms.
In doing HRED SS-RICS set of similar, in principle, lung algorithms. So makarom set individual functions, including long-term memory and perception, in the end will come in ready-cognitive system — an artificial mind.
What does this mean in practice? Here are a common example. Door — familiar to every detail of the interior, but in some countries it may function to do the mat. For a person to find a door, whether it is made of boards or blankets, but the robot must be programmed aggressively, otherwise it will simply ignore «incorrect» door and pass by.
Bot system SS-RICS will be many times to expel through rooms with different types of doors. As a result, it will remember the special features of doorways and learn to notice all the doors exotic types. With all of this up some qualities of human memory will reduce the computational load, and hence the cost, size and power consumption bots. Moreover, unlike people, to teach each individual mechanical fighter is not necessary — combat experience of individual machines will be an invaluable material for new copies of artificial intelligence.
Currently, experts at HRED develop three major multipurpose method: memory, language and perception (eg color recognition). If this work is completed successfully, the bots will be able to simply carry out simple orders of the commander, for example, «Scout it squared X», «Strip the building to the right», «blue car fired» etc. Practically brand new era comes combat vehicles not infirm radio controlled «toys», but real fighter teammates.