MIT Team Develops AI That Thinks and Acts Simultaneously

Robots can think and act simultaneously. An MIT research team has developed an artificial intelligence (AI) framework called 'VLASH' that allows robots to plan their next moves while in motion. Developed by Professor Song Han's team, this technology realizes seamless movement by eliminating 'computation time,' a bottleneck that previously slowed down robot reaction speeds.

Conventional AI models required robots to wait until calculations were complete, resulting in frequent stoppages. Conversely, moving while reasoning caused instability issues due to time lags. To overcome these limitations, the researchers proposed VLASH, which does not incur additional execution costs.

The researchers enabled the robot to plan its next action based on its 'future state' rather than its current position. The principle involves predicting the location when the current movement ends and simultaneously calculating the next action based on that data. The team achieved these results by applying 'action bundling' and 'learning augmentation' techniques.

In an experiment sorting colored cubes into boxes, VLASH improved operation speed by two to three times while maintaining the same accuracy as the state-of-the-art existing technology. The research paper has been published on arXiv, with detailed presentations scheduled for the 'IROS 2026' conference in Pittsburgh on September 27.