N-learning: An Approach for Learning and Teaching Skills in Multirobot Teams

Abstract

SummaryWe propose the N-learning practical approach for teaching and learning behaviors in a multirobot system, which is performed through mandatory behavior acquisition based on interactions between the robots at execution time. The proposed methodology can be used to self-program the robots of a team by programming only a single robot with a set of codes that contain behaviors to be transferred and used by other robots as necessary. These codes are implemented in a modular fashion. An advantage of our approach is that when a team of robots is required to perform a specific mission, the set of behaviors required to accomplish that mission can be implemented only once in a single robot or in a distributed fashion. Then, these distributed behaviors are transferred to each of the other robots in the team according to their demand, without the need to reprogram them by hand since the robots in the team can share them autonomously. As an application example, a human critic can teach (or program) only one or a few robots, and these robots are thus able to exchange knowledge with the other team members since they have been preinstalled to run the N-learning system basics. Simulated and real robot experiments are performed to demonstrate the feasibility and validation of our approach.

Online Learning and Teaching of Emergent Behaviors in Multi-Robot Teams

“…The first time that the transferring (learning and teaching) of pre-programmed behaviors was proposed was in the work of Costa et al [4], through the approach called N-learning. In the N-learning approach, behaviors are blocks of code with information about the execution of a specific maneuver or action, which can be shared throughout the multirobot team at execution time.…”

“…as one or more behaviors that enable the robot team to adapt to situations that are not previously taught in its initial programming. According to the authors [4], N-learning allows a robot to share one or more behaviors that it contains to another mate, which in turn eliminates the need for programming all robots within a team with the needed behaviors. These behaviors can be transferred in a distributed fashion using the interaction between the robots in a group at run time.…”

“…Our approach is not like machine learning approaches. As previously stated in [4], the N-Learning approach is a knowledge transference algorithm, not a machine learning algorithm. The process of learning is through the transference of blocks of code, and it does not use update processes to learn a behavior.…”

“…Hence, in the current work, we propose an extension of that previous work [4], in which the N-learning (now extended) is able to acquire new behaviors in execution time besides only using previously set behaviors as in that work, thus allowing the multirobot system (MRS) to eventually learn new, emergent behaviors. Therefore, we borrow some theory and implementations from the previous works, for the better understanding and implementation of the current work.…”

“…In all the above-mentioned problems, the use of N-learning is advantageous. A multirobot system heterogeneity can be used alongside N-learning by sharing different behaviors (or creating new ones) demanded by the raising of unexpected problems [4]. Most of the problems above can use solutions such as the Program by Demonstration (PbD) approach.…”