Researchers have built algorithms that aim to adapt the actions of robots to complex terrain conditions, and make them cooperate
How do you make robots perform human tasks? Artificial intelligence is a branch of computer science that studies the ability to program computers to act like the human brain. Automated Planning is a field of artificial intelligence that deals with building algorithms to produce plans - series of actions - to be carried out by autonomous agents such as robots in order to achieve goals.
"The goal of automatic planning is to program systems that can be controlled, that will perform tasks and achieve the results we wish for them. This is how, for example, the probe (space vehicle) of the space agency NASA works for Mars; It is based on a planning algorithm that produces a series of actions for it - so that it reaches the relevant places and performs the requested experiments," explains Prof. Ronan Brafman from the Department of Computer Science at Ben Gurion University, who, together with his team, is engaged in automatic planning and builds algorithms that robots can run in the future.
The main algorithms that build a series of automatic actions are based on heuristic functions; They lead the algorithm towards the goal and cut branches (possible solutions) that are not in its path. For example, if a programmer wants a robot to get from one point to another in a limited time, he can calculate with the help of these functions what is the shortest path and rule out options that make it longer. This way he manages to focus on the series of actions that will lead him to the goal efficiently.
But what happens when the situation is complex and unpredictable and it is not possible to know in advance what the desired actions and results are, such as in a disaster area that requires rescue and rescue missions? In such a case a pre-written series of actions will not work and a plan is required in which one action will lead to another, depending on the changing situation in the field. In their latest research, which won a research grant from the National Science Foundation, Prof. Brafman and his team built algorithms aimed at adapting the actions of robots to the field conditions and making them synchronize and cooperate when necessary even without communication between them. "The purpose of these algorithms is to make several robots perform actions in a complex situation (for example in a building destroyed by an earthquake) in order to achieve a certain goal (for example to evacuate the building quickly). But some operations, such as lifting the wounded and rescuing them, are difficult to perform independently and require cooperation. In addition, communication often breaks down in such areas, and in any case robots, like humans, can only receive partial information about the world and the environment (since their number of sensors is limited). The algorithms we built were designed to allow them to operate efficiently and cooperate even in such complex situations," explains Prof. Brafman.
The algorithms, which are based on software that receives a description of a goal and breaks it down into a set of actions, treat robots as a split body that works together and leads to the achievement of the goal in stages. In the first stage, a task and problem solving plan is built for each robot, assuming that there is immediate and optimal communication between all the robots. This program is actually a theoretical decision tree that describes what each robot will do based on the information that all the robots have from all the sensors together. In the second step, the algorithms analyze the decision tree and check what information is necessary for the plan to be implemented. They adjust the decision tree according to the findings and add sensing and communication operations to bridge the information differences between the robots in favor of executing the plan.
For example, the algorithms can divide the rooms of the destroyed building between the robots (that is, give each of them an area of responsibility), and when necessary make them go to other rooms and share information with other robots, according to the data of the area and the information they will pick up in their sensors; And the robots that receive the information will be able to act accordingly (for example, come to help).
"The idea is to produce robots that will make calculations and make decisions similar to humans or even better. That is, to build an automatic tool that can help humans or act in their place in dangerous areas (for example, those where a nuclear disaster occurred). They will be given a task and they will know how to break it down into the required actions in real time. The algorithms we build are the brains behind these robots. We try to improve them regularly, so that they will be more sophisticated and will be able to deal with models as detailed as possible", Prof. Brafman concludes.
Life itself:
Prof. Ronan Brafman, 57, married with three sons (21, 25, 28), lives in Moshav Ramot Meir. likes to do sports, ride a bike and travel.
More of the topic in Hayadan:
