Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Robot Object SLAM Using Distributed Variational Inference

Hanwen Cao, Sriram Shreedharan, Nikolay Atanasov

TL;DR

The paper tackles distributed multi-robot object SLAM by casting collaborative estimation as variational inference over a communication graph with consensus constraints on the object landmark state \\mathbf{y}. It develops a distributed mirror-descent algorithm with KL-regularization, yielding a Gaussian VI update that leads to a distributed MSCKF for object SLAM. The approach enables one-hop, fully distributed collaboration, achieving improved trajectory and object-map accuracy and favorable scaling compared to centralized methods, with modest communication and computation overhead. This framework provides a scalable, infrastructure-light path toward coordinated multi-robot tasks that rely on a common object map for planning and decision-making.

Abstract

Multi-robot simultaneous localization and mapping (SLAM) enables a robot team to achieve coordinated tasks by relying on a common map of the environment. Constructing a map by centralized processing of the robot observations is undesirable because it creates a single point of failure and requires pre-existing infrastructure and significant communication throughput. This paper formulates multi-robot object SLAM as a variational inference problem over a communication graph subject to consensus constraints on the object estimates maintained by different robots. To solve the problem, we develop a distributed mirror descent algorithm with regularization enforcing consensus among the communicating robots. Using Gaussian distributions in the algorithm, we also derive a distributed multi-state constraint Kalman filter (MSCKF) for multi-robot object SLAM. Experiments on real and simulated data show that our method improves the trajectory and object estimates, compared to individual-robot SLAM, while achieving better scaling to large robot teams, compared to centralized multi-robot SLAM.

Multi-Robot Object SLAM Using Distributed Variational Inference

TL;DR

The paper tackles distributed multi-robot object SLAM by casting collaborative estimation as variational inference over a communication graph with consensus constraints on the object landmark state \\mathbf{y}. It develops a distributed mirror-descent algorithm with KL-regularization, yielding a Gaussian VI update that leads to a distributed MSCKF for object SLAM. The approach enables one-hop, fully distributed collaboration, achieving improved trajectory and object-map accuracy and favorable scaling compared to centralized methods, with modest communication and computation overhead. This framework provides a scalable, infrastructure-light path toward coordinated multi-robot tasks that rely on a common object map for planning and decision-making.

Abstract

Multi-robot simultaneous localization and mapping (SLAM) enables a robot team to achieve coordinated tasks by relying on a common map of the environment. Constructing a map by centralized processing of the robot observations is undesirable because it creates a single point of failure and requires pre-existing infrastructure and significant communication throughput. This paper formulates multi-robot object SLAM as a variational inference problem over a communication graph subject to consensus constraints on the object estimates maintained by different robots. To solve the problem, we develop a distributed mirror descent algorithm with regularization enforcing consensus among the communicating robots. Using Gaussian distributions in the algorithm, we also derive a distributed multi-state constraint Kalman filter (MSCKF) for multi-robot object SLAM. Experiments on real and simulated data show that our method improves the trajectory and object estimates, compared to individual-robot SLAM, while achieving better scaling to large robot teams, compared to centralized multi-robot SLAM.
Paper Structure (30 sections, 3 theorems, 79 equations, 4 figures, 4 tables, 1 algorithm)

This paper contains 30 sections, 3 theorems, 79 equations, 4 figures, 4 tables, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Multi-robot factor graph optimization
  4. Multi-robot filtering
  5. PROBLEM STATEMENT
  6. Distributed Variational Inference
  7. Variational inference
  8. Distributed mirror descent
  9. Linear Gaussian case
  10. Distributed MSCKF
  11. State and observation description
  12. Consensus averaging
  13. State propagation
  14. State update
  15. Feature initialization
  16. ...and 15 more sections

Key Result

Proposition 1

The optimizers of eq:distributed_MD satisfy:

Figures (4)

  • Figure 1: Illustration of multi-robot object SLAM via distributed multi-state constraint Kalman filtering. The left images are inputs for the robots, where the red are geometric features (extracted by FAST fast_feature) and the green are object detections (by YOLOv6 yolov6). The geometric features and object bounding box centroids are used as observations. When common objects are observed by communicating robots, a consensus averaging step is performed to align the estimated robot trajectories and object positions.
  • Figure 2: Trajectory and object estimates of (a) 3 robots on KITTI sequences 00, 05, 06 and 08, (b) 15 robots in simulation.
  • Figure 3: Time consumed by different components per robot per frame, including MSCKF update, consensus averaging, and other (prediction and landmark initialization).
  • Figure 4: Analysis of the effect of the robot network connectivity. The dashed lines show RMSE without averaging.

Theorems & Definitions (5)

  • Proposition 1
  • proof
  • Proposition 2
  • Proposition 3
  • proof