Torobo, a Japanese humanoid robot, has reached new heights in autonomous cleaning thanks to a powerful learning method known as Deep Predictive Learning (DPL).
Tokyo Robotics just released a new demo of its flagship humanoid smoothly wiping a table on its own, guided by technology created at Waseda University’s Ogata Lab. The method enables Torobo to predict and adjust its movements in real time, bridging the gap betwen simple robot motions and complex, natural human actions.
As the robot moves, blue dots mark its current focus points, while red dots highlight the next movements it plans to make. This predictive approach allows Torobo to maintain smooth, coordinated wiping actions, applying just the right amount of pressure for effective cleaning. It learned this skill by watching human operators control it remotely, recording each demonstration’s joint angles, visual feedback, and pressure data.
After observing and recording these teleoperated sessions dozens of times, Torobo applied what it learned to replicate the motions autonomously.
Built for industrial work alongside humans, Torobo has arms with seven joints that enable it to move in human-like ways.
Tokyo Robotics has recently showcased other demonstrations of Torobo using its versatile arms to saw wood and accurately hammer nails.
Tokyo Robotics plans to launch a new version of Torobo in Japan by March 2025, with an international release slated for October.
