Advanced Material Science
| |
Advanced Material Science |
|
Regulations for master students |
Autonomous Driving of Nanocars The discussion will take place at O. Hofmann's webex server (click here) With the world’s first artificial-intelligence-controlled nanocar we are planning to participate in a nanocar race at the Center for Materials Development and Structure Studies (CEMES-CNRS) in Toulouse, France in 2021. At this race, participants have to direct a single molecule across a “race-track”, which is set on a metallic substrate, in order to control their nanocar via an STM-tip without being in physical contact with it. Although nanocars can be readily synthesized with different shapes and properties, the physics that govern the molecule’s movement is complex and involves the interaction between the molecule and the tip as well as the molecule and the substrate. Therefore, it is far from straightforward for humans to manoeuvre the nanocar and predict the result of a performed action. To improve the performance, an artificial intelligence based on reinforcement learning is implemented, which can perform various actions even in subsequently changing environmental conditions. The AI is implemented in the form of an off-policy reinforcement learning algorithm, known as Q-Learning algorithm. Being off-policy allows the AI to learn from human generated data but also from otherwise generated data. This means, that the AI can be trained without operating directly at the STM. After training, the AI is capable of driving the nanocar by controlling the STM-tip position and the applied voltage based on the orientation and position of the nanocar on the surface. This knowledge of molecular manipulation can be used for developing future bottom-up constructions of nanotechnology. |