Skip to content

DynoSLAM: Dynamic SLAM with Generative Graph Neural Networks for Real-World Social Navigation

Home / Papers / DynoSLAM: Dynamic SLAM with Generative Graph Neural Networks for Real-World Social Navigation

DynoSLAM: Dynamic SLAM with Generative Graph Neural Networks for Real-World Social Navigation

Authors: Danil Tokhchukov, Veronika Morozova, Gonzalo Ferrer Date: 2026-05-04 Paper ID: openalex:2605.02759

Summary

DynoSLAM is a dynamic SLAM architecture designed for navigation in crowded environments by relaxing the static-world assumption. It integrates socially-aware GNNs directly into factor graph optimization, treating pedestrian motion as a stochastic process rather than a deterministic heuristic. By utilizing Monte Carlo rollouts to derive mean and covariance for future states, the model provides a probabilistic safety envelope that significantly improves tracking robustness compared to traditional approaches.

Key Contributions

  • Introduces DynoSLAM, a dynamic SLAM framework that integrates socially-aware GNNs into factor graph optimization.
  • Formulates pedestrian motion forecasting as a stochastic world model to capture multimodal uncertainty.
  • Implements a dynamic Mahalanobis distance factor to embed epistemic uncertainty into the optimization, mitigating the ‘argmax’ failure mode in tracking.

Open Questions & Future Work

Key Concepts

  • dynamic-graphslam: A SLAM framework that incorporates socially-aware GNNs for motion forecasting directly into factor graph optimization.

Archivist Review

I approved the ‘Dynamic GraphSLAM’ concept because it introduces a novel, reusable mechanism for integrating neural motion priors into factor graph-based SLAM, which is central to the paper’s contribution. The open question regarding ‘Generative Flow Matching for Motion Prediction’ was also approved as it identifies a clear, theoretically grounded path for improving uncertainty estimation beyond current sampling-based surrogate methods in dynamic robotics.

Approved Concepts

  • Dynamic GraphSLAM: It provides a way to incorporate stochastic, interaction-aware neural priors directly into SLAM optimization frameworks, moving beyond rigid heuristic motion models in dynamic environments.

Approved Open Questions

  • Generative Flow Matching for Motion Prediction: This shift from surrogate noise to learned generative dynamics is essential for improving the accuracy and theoretical grounding of uncertainty estimation in complex human-robot interaction scenarios.

Metadata & Links

url
https://arxiv.org/abs/2605.02759
paper_id
2605.02759
paper_source
openalex
domain
computer-vision
tags
computer-visionrobotics
architectures
datasets
concept_slugs
dynamic-graphslam
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-05-07T06:05:14Z