Home Physics AI Learns to Play with a Slinky

AI Learns to Play with a Slinky

AI Learns to Play with a Slinky


• Physics 16, s119

A brand new synthetic intelligence algorithm can mannequin the conduct of a set of objects, resembling helical springs or pendulums, utilizing a way that may extrapolate to things that the algorithm hasn’t beforehand analyzed.

Wirestock Creators/inventory.adobe.com

Like many professions, physicists are on the lookout for methods to make use of synthetic intelligence (AI) to enhance job efficiency (with out hopefully being changed by it). In that vein, researchers have now developed an AI algorithm that may analyze the movement of a set of objects, resembling swinging pendulums or bouncing Slinkies, after which use that info to develop a normal mannequin of the forces appearing on these methods [1]. This technique can extrapolate to things that weren’t beforehand examined 100 instances quicker than different AI strategies that don’t attempt to generalize.

Physicists have beforehand proven that AI algorithms can routinely uncover hidden relationships in advanced knowledge (see Viewpoint: Physics Insights from Neural Networks). Nonetheless, a typical criticism of those algorithms is that they’re too particular in that they aim just one system, says Qiaofeng Li from the College of California, Los Angeles. Li and colleagues have developed a way that may be taught from a set of circumstances and may then generalize to others.

The researchers supplied their algorithm with a set of trajectories, such because the positions and velocities of falling Slinkies—every having a distinct stage of stiffness. The algorithm analyzed these trajectories after which constructed a normal mannequin with adjustable parameters that allowed it to investigate in about one minute the movement of any Slinky, even one which was not in its coaching set. For less complicated methods, resembling pendulums and oscillating electrical circuits, the evaluation might take as little as a number of seconds. The staff foresees making use of the strategy to the mechanical evaluation of organic cells in several mediums or the management of robots in quickly altering environments.

–Michael Schirber

Michael Schirber is a Corresponding Editor for Physics Journal primarily based in Lyon, France.


  1. Q. Li et al., “Metalearning generalizable dynamics from trajectories,” Phys. Rev. Lett. 131, 067301 (2023).

Topic Areas

Computational PhysicsMechanics

Associated Articles

Magnetic Microdisks Don’t Always Reciprocate
Performance Capacity of a Complex Neural Network
Controlling the Bounce of a Bottle

Extra Articles



Please enter your comment!
Please enter your name here