Distribution Estimation for Global Data Association Via Approximate Bayesian Inference
Yixuan Jia, Mason B. Peterson, Qingyuan Li, Yulun Tian, Jonathan How
AI summary
Problem
Existing global data association methods typically return a single most-likely solution, which fails in ambiguous environments with repetitive or symmetric structures where multiple valid associations coexist.
Approach
The authors reformulate data association as sampling from a posterior distribution and use Stein Variational Gradient Descent and Langevin dynamics to evolve a particle set that collectively covers multiple solution modes.
Key results
- Captures highly peaked and uniform solution distributions in ambiguous scenarios
- Outperforms probabilistic ICP variants on highly non-uniform distributions
- Enables efficient GPU-parallelizable computation without manual kernel tuning
- Validates accurate distribution estimation on simulated object maps and real-world point cloud registration
Why it matters
It equips robots with reliable uncertainty quantification for perception and localization in symmetric environments, preventing catastrophic errors in downstream mapping and navigation.
Abstract
Global data association is an essential prerequisite for robot operation in environments seen at different times or by different robots. Repetitive or symmetric data creates significant challenges for existing methods, which typically rely on maximum likelihood estimation or maximum consensus to produce a single set of associations. However, in these ambiguous scenarios, the distribution of solutions to global data association problems is often highly multimodal, and such single-solution approaches frequently fail. In this work, we introduce a data association framework that leverages approx- imate Bayesian inference to capture multiple solution modes to the data association problem, thereby avoiding premature commitment to a single solution under ambiguity. Our approach represents hypothetical solutions as particles that evolve via deterministic or randomized updates, naturally parallelizable on GPUs, to cover the modes of the underlying solution distribution. Simulated and real-world experiments with highly ambiguous data show that our method correctly estimates the distribution over transformations when registering point clouds or object maps. Code is available at: https://github.com/ mit-acl/mmda.