Chimera: enabling hierarchy based multi-objective optimization for self-driving laboratories.

Area of Science:

• Decision-making and Optimization
• Computational Science
• Systems Engineering

Background:

• Multi-target optimization is complex, especially when experiments or computations are costly, necessitating minimal evaluations.
• Existing methods often require extensive prior knowledge of individual objectives, limiting their applicability.

Purpose of the Study:

• To introduce Chimera, a versatile achievement scalarizing function for multi-target optimization.
• To address the challenge of limited evaluations in complex decision-making processes.
• To provide a method applicable to any set of unknown objectives without requiring detailed prior knowledge.

Main Methods:

• Chimera integrates a priori scalarizing and lexicographic approaches.
• It is demonstrated with various single-objective optimization algorithms on analytic multi-objective benchmarks.
• Practical applications include robotic sampling auto-calibration and inverse-design of excitonic systems.

Main Results:

• Chimera enables a broad range of optimization algorithms to quickly identify ideal conditions.
• The function proved effective in optimizing a virtual robotic sampling sequence.
• It facilitated the inverse-design of a four-pigment excitonic system for efficient energy transport.

Conclusions:

• Chimera is a powerful tool for efficient multi-target optimization, particularly when evaluations are limited.
• Its applicability is shown across diverse scientific and engineering domains.
• The interpretability of Chimera aids in validating design choices and tailoring system parameters.