The goal of the XploreR project was to apply autonomous exploration using a single robot. In the next step, we combined the autonomous exploration algorithm implemented in a mobile robot with a virtual reality environment for visualizing collected data.
Depth estimation plays a big role in robotics, autonomous driving (mapping, localization and obstacle avoidance), augmented reality and many other computer vision technologies across many
Our recent migration of Karelics Brain from ROS 1 to ROS 2 had our whole team learning a lot of new things in a relatively short time period.
It is then obvious that when we have heard that a new installment of the RosDevDay conference will also take place in 2021 as an online event, in spite of all the drawbacks imposed by the ongoing COVID pandemic, we have decided that Karelics also had to be there and not by just simple attendance but also by being one of the gold sponsors of the conference.
In our latest blogpost we have shown that we have already successfully finished, tested and shown to the world the first version of Karelics Brain.
At Karelics, one of our main missions is to shed more light on the fact that the 21st century is already shaping up to be the century of robotics and is high time we already made the general public aware to the existence of intelligent robots that can be our day to day helpers and can make our lives better and easier.
The world is highly complex. To operate in this complex world with self-driving robots, it would be hard to program all the tasks by hand, creating rules for every subtask and action. We humans can complete really complex tasks just by setting for ourselves the final goal for what we are supposed to do.
The first question that you face when building a Robot is what should be the best size for it. There are several ways to do it:
Last year, I got an incredible opportunity to join a team at University of Eastern Finland and compete in MineRL competition. The competition sponsored by Microsoft aimed to push the boundaries of state-of-the-art reinforcement learning and sample efficient learning.
Can you train a deep learning agent in video game and then transfer this learned information to a real world robot? How can we handle the differences in visuals and action spaces between the game and reality?