PRISM helps robots move better by learning how forces and motion interact
Mechanical EngineeringRoboticsPhysics
THE AI ANGLE
Learning multiplicative physical interactions for robotic motor controlResearchers developed PRISM, an input component that lets neural networks learn multiplicative physical interactions directly from robot sensor data. The approach improved humanoid walking stability and robotic arm manipulation over standard networks without requiring larger models or explicit equations. Engineers can add this component to existing robot learning systems to maintain performance across varied loads and friction conditions.
THE TEACHING ANGLE
Students can debate whether robot policies benefit more from increasing general model size or structuring inputs to reflect physical mechanics.Read the original at techxplore.com Generate teaching or study materials
More in Mechanical Engineering
- US Foods to test inventory-scanning robots at 6 more warehousesSupply Chain Dive · November 18, 2026
- Single-propeller drone uses its own spin to scan surroundings in all directionsPhys.org — Technology · November 18, 2026
- Electron fingerprints help AI predict complex alloy properties with little training dataPhys.org — Technology · November 18, 2026
- A shared signal for humans and robots: Flexible display uses both color and near-infrared to communicatePhys.org — Technology · November 17, 2026
- Boston Dynamics’ Atlas humanoid lands first non-automotive trialThe Robot Report · November 17, 2026