How training robots to face their environment can be useful
Robots we develop are not granted with accuracy to do high precision tasks already. Exceptionally when you plan to face wild environments, one should make sure everything is clear about sensing and control/locomotion to avoid damages.
As we know already, training is the best way for us humans to improve and get outstanding results in a given amount of time, similarly, robots need to practice first and be observed while they train on a specific environment. In this first article, I will talk about estimation and learning and how it could help with our actual common robots. We shall not enter too much on the details of technologies but focus on their help.
Autonomous Aerial Drones :
Drone technology when used for indoors tasks need to learn it’s environment first to be dynamic lately, the process of learning it’s environment to a drone cares a lot about building maps as they move.this is all the concern of SLAM: simultaneous localization and mapping
As long as a map is available the robot can move very smoothly in his environment and afford computational motion planning algorithms(discussed in a later article) autonomously to perform any indoor task such as survey and tracking.
It appears to be easy but making a UAV so dynamic to slam requires to switch between different behaviors that can be achieved through hybrid automata to make sure it doesn’t collide while learning it’s the environment. we shall go in further explanations with Autonomous Terrestrial drones
Autonomous Terrestrial drones:
There is a lot to say about this mobile robotics technologies, but as our topic is based on the estimation and learning by these robots given a wild environment, I shall be brief.
These vehicles exist in several configurations that can perform indoors ou outdoors applications. Manned control is quite easy to achieve, but as discussed earlier, the autonomous or unmanned control of this robot is a major robotic issue.
One of these robots was sent to Mars named curiosity to give information about that unknown and unexplored environment.
But Let’s come back on earth first
This is not essentially an example of the common robots we use in our homes or for our orthodox activities.
we’ll discuss more the robots we do use in our homes to improve and benefit from them.
The 4mob Terrestrial Drone by STERELA
is indeed one terrestrial and cooperative robot you would like to work with, actually used for aircraft inspection and heavy load transport agricultural tasks, the only limit is your imagination.
No matter the task you need to perform the estimation and learning algorithms will help you to make to robot gets dynamic in wild environment such as the one presented in this picture
How can I navigate efficiently in such environment?
Good question, here’s the answer:
Learning the environment
Build Maps :The challenge is this environment is the variability of obstacles on the way, so if it is possible to have a precise map of a given space of work the robot needs to navigate on, it could be more easy to locate and be dynamic in this wild and constrained environment, once more, SLAM is a very profitable technology that will help us to build those precise maps we need.
this is an example of SLAM we can get with a mobile robot without using a camera(a case we saw earlier), just lidar measurements.
Why won’t we go ahead and get a better map?
well, there’s no problem with that, this is the result of countless measurements on which I worked on.
but if you have enough battery and you have the required sensors (lidars for example) you could get a full map of your environment enabling you to begin trajectory tracking that will not be discussed yet.
Here is a more complete occupancy grid map that can be implemented using MatLab or python.
we can see obstacles on the map and plan a safe trajectory, as shown by the green line.
So coming back to our navigation problem in wild the environments.
we just need to use the appropriate sensors that could describe what you consider an obstacle for your robot and showcase it on your map. Then Explore
STATIC ROBOTS: scara robot
I can’t end without talking about scara robots, This is indeed one of the most used robots in the world, it performs high-speed pick and place tasks with a precision that cannot be done by numbers of men. This is one of the best examples to illustrate the learning algorithms.
This robot applies object detection algorithms that require skills in image processing and probabilistic models to make the pick and place robust to track only it’s a target during the task and successfully place it at the desired position.
This is the result of a basic image processing algorithm with MatLab that helps us to target an object given a whole workspace.
we could even recognize and label different objects in a given environment depends on the form we search. focus and reason on them, once more we use basic image processing
The picking process could deeply require probabilistic models depending on the task that leads the pick. For example, extracting objects with defects on a given sequence of a production line. The robot needs once more to learn how to characterize an object in good shape withdeeper skills of image processing that may call on machine learning algorithms.
This is an example of a scara robot thatfocuses recognition on the form of the object and the word written through Character recognition, these two are discussed topics of image processing and machine learning.
These technologies we use: image processing, SLAM, machine learning are all used in robotics applications to achieve speed, precision, and flexibility while performing different tasks Autonomously in a wild or constrained environment.