NATURE BIOTECHNOLOGY VOLUME 35 NUMBER 4 APRIL 2017 311 needs to be developed for describing experimental protocols, operating system assemblies, and recording results. Currently, Maholo’s proprietary software system and application program interface enables only Maholo to perform experimental tasks. Although robotic operation frameworks are also provided by several services, including Transcriptic, Emerald Cloud Lab, Aquarium (a product of Eric Klaven’s group at the University of Washington, Seattle) and Synthace’s (London) Antha, no practical framework has been proposed to transfer protocols among different laboratory automation systems or to freely integrate different tools and automation systems to automate complex experimental operations. We propose that a scalable laboratory automation should be split into two layers: first, a standard semantic layer, or process ontology layer, in which experimental workflows are described by connecting various job processes (‘process description’); and second, a translation layer, in which each job process described in the standard semantics is compiled into robotic operations for a given operation environment (‘process-tooperation mapping’). The first layer could be actualized from an open science community by harnessing previous efforts for process ontology for laboratory experiments6–9. The second layer would require the manufacturing side to prepare its process-to-operation mapping for different job processes defined using the standard syntax. At present, laboratory automation processes generally suffer from the ‘hard coding’of systems and manual adjustments to sensitive environmental differences, which are usually not well documented. However, for LabDroids that have sensing systems and form an RCBL, robotic behaviors in a specified environment could be fed back to the central computing system and automatically optimized with the support of artificial intelligence. This would allow the ideal automated processto-operation mapping or ‘real world programming of experiments’(much like compiling a program script in a particular computational hardware environment). Phase I of the Robotic Biology Consortium is to build a few small-scale RCBLs by early 2020. Each RCBL will be composed of multiple LabDroids, laboratory automation systems, and human-usable experimental tools and equipment. We plan to demonstrate fully remote operation of complex experiments in genomics, proteomics, and high-content cell screening, and to showcase the reproducibility of the experiments exchanged between different RCBLs. job-specific laboratory automation systems combined with human-usable tools. In such a robotic crowd (or cloud) biology laboratory (RCBL), experimental protocols are sent online and samples are shipped from all over the world (Fig. 2b). Reagents and samples are barcoded and linked to the corresponding process in the protocols. Barcode scanning and automated internal delivery systems allocate reagents and samples around LabDroids that operate experiments. A timeline of each experimental step is extracted automatically from the submitted protocol, with information for materials and instruments. A single LabDroid does not perform a whole experiment by itself, but a team of LabDroids can perform multiple experiments simultaneously based on an ‘agent crowd operation’. For each sub-process in the different experiments in the queue, the central computing system dynamically assigns available LabDroids that are close to optimal instruments and thereby maximizes the production of the whole RCBL. The development of an RCBL could provide a new …