N-Learning, uma abordagem para ensino e aprendizagem autônomos em sistemas multirrobôs
We propose the N-Learning paradigm, which allows the sharing (teaching and learning) of behaviors in multi-robot systems autonomously and at run time. The proposed paradigm is based on behavioral robotics and uses cooperative learning. In the formal model, robot behaviors are represented in the fo...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | doctoralThesis |
| Language: | pt_BR |
| Published: |
Brasil
|
| Subjects: | |
| Online Access: | https://repositorio.ufrn.br/jspui/handle/123456789/26936 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | We propose the N-Learning paradigm, which allows the sharing (teaching and learning) of behaviors in multi-robot systems autonomously and at run time. The proposed
paradigm is based on behavioral robotics and uses cooperative learning. In the formal
model, robot behaviors are represented in the form of a graph, where complex behaviors
can be broken down into simple behaviors that, in turn, can be performed simultaneously.
N-Learning allows to change the scope domain of the robot without the need for reprogramming. That is, a robot that does not have compatible behaviors for a given domain
can change and learn from the other robots acting in that domain. This feature is useful
when there are a large number of robots and several, different missions (in different domains) to be fulfilled. N-Learning can also be used with emerging behaviors that need to
be shared with the team. To validate the paradigm, a reference implementation was developed based on the Python language and the Robot Operating System, using the Stage
simulator and real robots. Results show that individuals in a group of robots can learn
through interaction in the multirobot system. The team comes from a state of less knowledge of robots, individually (ie, robots possessing ability to execute a few behaviors) to
a state of more knowledge (robots accomplishing more behaviors, learned online). With
this approach, behaviors that are specific to certain environments, already existing, do
not need to be preprogrammed in the robots, which can learn them with the other robots
of the team. The experiments demonstrate the versatility of N-Learning, validating our
approach. |
|---|