Learning Adaptive Solvers for Distributed Factor Graph Optimization on Matrix Lie Groups
2026-07-09 • Robotics
Robotics
AI summaryⓘ
The authors developed DeepCORD, a new method to help multiple robots or systems work together more efficiently when solving complex geometry problems. Traditional approaches need a lot of manual tweaking and mostly focus on simple robot positions, but DeepCORD uses a learning-based system that adjusts itself automatically. It works well even when communication between robots is slow or irregular. Their tests show that DeepCORD performs better than existing methods in different realistic scenarios involving 3D robot positioning and aligning projective maps.
distributed optimizationfactor graphmatrix Lie groupsRiemannian optimizationSE(3) pose graphSL(4) projective alignmentself-supervised learningparallel computingrobotic perceptionasynchronous communication
Authors
Jaeho Shin, Maani Ghaffari, Yulun Tian
Abstract
Modern robotic perception increasingly involves large-scale geometric optimization problems distributed across multiple robots or sessions. However, existing distributed solvers often depend on brittle hand tuning and primarily target rigid body pose graphs. To address this, we present DeepCORD, a learning-augmented framework for distributed factor graph optimization on general matrix Lie groups. By unfolding a parallel and accelerated Riemannian optimizer into differentiable iterations, DeepCORD learns a self-supervised feedback policy that dynamically adapts solver parameters according to the optimization phase and communication status. The resulting method enables adaptive distributed optimization over matrix Lie groups under both synchronous and asynchronous communication regimes. Extensive experiments on real-world $\mathrm{SE}$(3) pose graph optimization and $\mathrm{SL}$(4) projective submap alignment show that our method achieves lower objective values than existing distributed baselines on most benchmarks across realistic operating scenarios.