It learned to walk and to carry — separately: a humanoid robot stitches two skills into one job

Teach a robot to walk in one room and to pick things up in another, and it usually cannot combine the two on a factory floor. A team at Seoul National University, led by professor Songhwai Oh, has built a way around that: RISE, an AI framework that learns from walking demonstrations and object-handling demonstrations collected separately, then performs multistep jobs in sequence — walk to the shelf, pick up the item, carry it on. The researchers validated the system in simulation and on a real humanoid robot, the Unitree G1. The work has been accepted to IROS 2026, a leading robotics conference. Robots that can chain their own skills together need far less hand-holding — and far fewer bespoke lessons for every new task.

SDG 9 · Industry, innovation and infrastructure — Mechanism: tech innovation