Research agent improves dye removal across 15 autonomous laboratory rounds

Researchers report in a preprint that their Embodied Science Agent improved wastewater treatment results over 15 continuous rounds of autonomous experiments. Final solution transmittance rose from 82% to 99.7%, corresponding to a 98.5% reduction in residual dye concentration. The system combines memory-based reasoning and Bayesian optimisation, which guides the search for experimental conditions, with learned policies for controlling laboratory instruments. It navigated over 200 million possible conditions, and its instrument-control policies generalised across varying instrument placements. The demonstration establishes results in wastewater treatment experiments; it does not establish performance across other scientific workflows.