Back to News
Source: Techxplore
Published November 11, 2026Source checked November 15, 2026

PRISM helps robots move better by learning how forces and motion interact

Read the article and explore its original source.

Checking audio availability…

Read original source
ShareLinkedInX
No image available

Quick brief

  • University of Michigan Engineering’s prism component aims to improve humanoid control by learning force–motion interactions directly from observations and commands.

What is confirmed

In the prism study from University of Michigan Engineering, PRISM (Polynomial Representations for Interaction-Structured Motor Control) was tested on humanoid locomotion and robotic manipulation, and the researchers reported higher simulated walking completion than a conventional MLP across 24-second trials.

The prism results explicitly include simulated humanoid walking, where prism learns multiplicative interactions among joint positions, velocities, past actions and desired walking speeds to support robot motor control without manually specifying physical equations.

In prism simulated walking tests, a prism-controlled humanoid completed 92.5% of 24-second trials without falling, compared with 51% for a conventional MLP, while a parameter-matched conventional network reached 52.3%, according to the report.

Evidence and editorial ownershipRequest a correction
Topics
robot

Original source

Techxplore · techxplore.com

Published November 11, 2026 · Source checked November 15, 2026

Weekly Robotics Brief

New robots, industry moves, and funding in your weekly robotics briefing.

Optional topics

Unsubscribe anytime.