The Missing Piece of Automated Scientist: Enabling lab autonomy with reward functions and executable workflows

Autonomous scientists enabled by large language models and agentic reasoning have become a compelling narrative for accelerating research. Here, we pose that laboratory autonomy is not primarily a text problem, since experiments have to be run in the world of instrument-addressable actions and constraints, and planned based on costs, latencies, and rewards. We argue for a bottomup foundation for self-driving laboratories (SDLs) that complements language-first planning by introducing frameworks that translate hypotheses into decisions executable on real tools. We propose three operational principles, including (a) adopt an information-efficiency objective that maximizes decision-relevant information per unit experimental cost through comprehensive acquisition and minimal decisive experimental design, (b) require hypotheses to compile into experimental footprints defined by experimentally-accessible actions, constraints, observables, and decision rules, and (c) treat reward design as the central bottleneck in closed-loop experimentation, especially when mapping high-dimensional measurements to utility and extending to sequential decision-making. We frame SDLs as long-horizon, out-of-distribution discovery systems where accelerating the full experiment–model–decision loop and emphasizing physical and chemical knowledge discovery over short-term correlations is necessary to translate lab-scale optimization into durable technological and manufacturing impact.