More information:
Master thesis: An MPC Framework For Planning Safe & Trustworthy Robot Motions
NEWS, Research, Robotics, Artificial Intelligence |
In chemical or biological laboratories, hazardous substances are regularly used. It is important that robots know where such substances are to be able to move safely in the same workspace and not turn over any test tubes or other containers. To do this, we need to digitize the environment. The challenge now is to represent the laboratory only as detailed as necessary and not as detailed as possible. Because for the motions to be safe and precise, a frequency of one kilohertz is required, i.e., a thousand times per second. This is exactly the rhythm of the robot, in which joint angles, distances, and forces are measured. A high-resolution image of the environment would slow the robot down, as it would have to process a lot of additional information within a very short time. Therefore, we simplify or abstract the detailed image of our lab with simple geometric shapes. This way we "relieve" the robot and it can keep the virtual representation of the environment in mind.
It is not that simple. Unfortunately, we don't get the virtual lab delivered. It would be ideal if the architect who designs a laboratory were to supply all the cabinets, tables, chairs, and measuring instruments in a three-dimensional image. Currently, we still create this image ourselves by positioning each object individually in our so-called environment replica. This is manual work. However, the simplification for fast processing in the robot control loop then works automatically in our approach. After that, precise distances between all objects are calculated and safety distances are also built in. Incidentally, in the next step, we plan to continuously regenerate and update the virtual lab using live images from cameras that are accurate to the centimeter.
"My" topic is experiment design. Chemical reactions depend on many parameters - temperature, concentrations of added substances, catalysts added, etc. This opens a large parameter space over which I optimize these reactions as part of my doctoral research. Therefore, I am working on algorithms that learn from experiments and suggest improvements for future ones. By the way, such topics will also play a role in the future for the first AI experimentation space at TUM. The fact that we have taken up our collision avoidance idea also has to do with the fact that, in my master's thesis, I dealt with optimal motion planning for robots that maintain a distance from humans that is perceived as more comfortable. I published a paper on this last year at ICRA, the other major robotics conference.
More information:
Master thesis: An MPC Framework For Planning Safe & Trustworthy Robot Motions
Receive our featured stories, news, events, and videos in your inbox every month, subscribe here!