New AI technique helps robots complete tasks twice as fast by letting them 'think ahead'

Scientists have created a fresh AI technique called VLASH, which can accelerate robots by up to 11.8 times without needing more computing power. Most robots, controlled by vision-language-action models, perform erratically - they reach, pause, and then adjust their movements. The main issue is the stop-and-go rhythm caused by traditional VLA models: after completing one set of instructions, the robot waits for the model to calculate the next set.

This delay hinders the use of VLA-controlled robots for tasks demanding continuous, real-time interaction. The VLASH system aims to eliminate this waiting period by forecasting the robot's next actions while it finishes its current ones. In tests, robots using VLASH completed tasks 1.5 to 2 times faster while keeping most or all of their accuracy.

Maximum reaction latency dropped by up to 11.8 times, depending on the computer hardware. Researchers from MIT, Nvidia, Caltech, UC Berkeley, UC San Diego, and Tsinghua University in China described the system in a paper on the arXiv preprint server and will present it at the Intelligent Robots and Systems Conference this fall. VLASH essentially acts as the brain of advanced robots.

It combines camera images with human instructions and the robot's current state, then translates this information into physical movements. VLASH predicts the robot's future state, not the entire environment. This prediction could eventually be combined with more advanced world models. However, researchers noted that faster reactions do not guarantee improved safety around humans.

The real test would be putting these robots in human-centered environments, like search and rescue operations. VLASH has not been tested in challenging conditions like poor lighting, smoke, unstable terrain, or with damaged cameras and unreliable communications. Further research is needed to determine VLASH's real-world impact and safety.