Research & Papers

Julia's MixedComplementarityProblems.jl beats closed-source PATH

Open-source solver clears 100x more trajectory games than PATH in benchmarks

Deep Dive

David Fridovich-Keil has open-sourced MixedComplementarityProblems.jl, a high-performance interior point solver for mixed complementarity problems (MCPs) written in pure Julia. This solver directly competes with PATH, the dominant closed-source alternative used in robotics and game theory, offering identical reliability on standard benchmarks while dramatically improving throughput.

The key innovation is native support for batched, parallel processing across both CPU threads and NVIDIA GPUs. In benchmarks modeling multi-agent lane-change trajectory games—common in robotics planning—a CPU-multithreaded version of the solver processed batches 100x faster than sequential PATH calls. While the GPU backend also outperformed PATH, it only surpassed the multithreaded CPU on large per-instance KKT systems, revealing a regime-dependent performance characteristic worth considering for deployment scenarios.

Key Points
  • MixedComplementarityProblems.jl is an open-source interior point solver for MCPs, matching PATH's reliability on standard benchmarks
  • CPU-multithreaded version processes batches ~100x faster than PATH on multi-agent trajectory optimization benchmarks
  • GPU backend supports unmodified solver code but excels only on large KKT systems, revealing performance regime dependencies

Why It Matters

Delivers high-performance, open-source optimization for robotics and multi-agent systems, breaking closed-source PATH's monopoly

📬 Get the top 10 AI stories daily