EvoMO documentation

EvoMO is a PyTorch library for tensorized, GPU-accelerated multiobjective evolutionary optimization, built on EvoX. It represents populations, objective values, and evolutionary operations as tensors for numerical optimization, constrained optimization, and policy search in robot control tasks.

EvoMO is a sister project in the EvoX family, and shares its logo with EvoX.

This guide covers the current PyTorch implementation. Start with installation and the quickstart. If you already use EvoX, see workflows and the API reference.

English is the primary documentation language. A complete 简体中文 guide is also available. Use the language switch on each page to open the corresponding chapter; generated API documentation is shared in English.

What you can do

• Search for trade-off solutions with NSGA-II, NSGA-III, MOEA/D, RVEA, and other algorithms.

• Evaluate algorithms on DTLZ, ZDT, WFG, MaF, and LSMOP numerical benchmarks.

• Use constraint violations returned by constrained problems in constraint-aware selection.

• Assess solution sets with GD, IGD, and Monte Carlo hypervolume estimates.

• Integrate EvoX monitors and model parameter adapters into EvoMO workflows.

• Search for robot control policies with MoRobtrol after installing optional dependencies.

GPU performance depends on population size, objective count, algorithm, evaluation cost, and the runtime environment. Algorithms do not all support the same combinations of constraints, compilation, and vmap; check their interfaces and the relevant guides.

Getting started

• Introduction to EvoMO
• Installation and environment
• Quickstart: NSGA-II on DTLZ2

User guides

• Algorithms
• Benchmarks and custom problems
• Workflows and monitors
• Constrained multiobjective optimization
• Solution quality metrics
• Devices, compilation, and performance
• Neuroevolution and MoRobtrol
• Troubleshooting

Development and reference

• Development and contributing
• API reference
  • Algorithm parameter guide
  • Publications and citation
  • Algorithms
  • Numerical problems
  • Constrained problems
  • Neuroevolution
  • Workflows
  • Selection operators
  • Metrics
  • Utilities
• Non-dominated sorting backends

Project and citation

Source code · Issue tracker · EvoX documentation

For related algorithm papers and downloadable BibTeX entries, see Publications and citation.

When using EvoMO in research, please cite:

@article{evomo,
  title = {Bridging Evolutionary Multiobjective Optimization and {GPU} Acceleration via Tensorization},
  author = {Liang, Zhenyu and Li, Hao and Yu, Naiwei and Sun, Kebin and Cheng, Ran},
  journal = {IEEE Transactions on Evolutionary Computation},
  year = 2025,
  doi = {10.1109/TEVC.2025.3555605}
}