An artificial intelligence agent can devise detailed instructions for a robotic arm and test them in a virtual laboratory. Credit: Jiaru Bai and Yuchi Zhao, University of Toronto
Artificial intelligence agents could help experimental robots overcome their narrow skill sets and limited mobility by planning and testing laboratory procedures before carrying them out. In a described system, a scientist provides the agent with the research goals and laboratory layout; the agent then creates a step-by-step protocol, builds a virtual version of the lab to test it and instructs a robot to perform the experiment in the real world.
Research robots can work continuously without tiring, but they generally have restricted capabilities and cannot easily coordinate without getting in one another’s way. AI agents could also use virtual laboratories to determine where equipment and robots should be placed, or to design an entire laboratory from scratch.
Robot chemist sparks uproar by claiming to have created new material
The system is described in a preprint posted on arXiv in October.1 It has not been peer reviewed.
During testing, the agent generated more than 200 commands directing a robotic arm to move between laboratory equipment and interact with it. It first simulated the procedure in a virtual laboratory to check that the steps worked smoothly. The real robotic arm then carried out the procedure to measure the optical properties of nanocrystals.
Evan Collins, a bioengineer at the Massachusetts Institute of Technology in Cambridge, described the concept as “very cool” and said it could substantially reduce the work needed to set up automated laboratories. However, he noted that adding the cameras and sensors required for an agent to direct experiments effectively can require more resources than setting up a laboratory manually.
Collision avoidance
Virtual simulations can make it easier to plan a robot’s movements and prevent collisions with other objects in the laboratory, Collins said.
Setting up an automated laboratory is often a time-consuming process involving trial and error, followed by extensive testing and troubleshooting. The system could simplify that process. Yuchi (Alan) Zhao, a roboticist at the University of Toronto in Canada and a co-author of the preprint, said researchers would not need to “babysit the system” once it had been built because the agent handles the planning and testing in a simulated environment.
Source: www.nature.com
