Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online Calibration of a Single-Track Ground Vehicle Dynamics Model by Tight Fusion with Visual-Inertial Odometry

Haolong Li, Joerg Stueckler

TL;DR

ST-VIO tightly fuses a differentiable, singularity-free single-track dynamics model with visual–inertial odometry to enable online online calibration of vehicle dynamics parameters while estimating VIO states. By embedding the dynamics as a multistep RK-integrated factor within a sliding-window graph and enforcing planar and geometry constraints, the method achieves improved tracking accuracy and significantly better forward-prediction under new control inputs. The approach adapts to changes in terrain and wheel properties, demonstrated across indoor/outdoor scenarios, stop-and-go maneuvers, and wheel swaps, while maintaining real-time performance. This work advances capability for model-predictive control and navigation planning in ground robots by providing online, data-driven adaptation of the vehicle dynamics model within the VIO framework.

Abstract

Wheeled mobile robots need the ability to estimate their motion and the effect of their control actions for navigation planning. In this paper, we present ST-VIO, a novel approach which tightly fuses a single-track dynamics model for wheeled ground vehicles with visual inertial odometry (VIO). Our method calibrates and adapts the dynamics model online to improve the accuracy of forward prediction conditioned on future control inputs. The single-track dynamics model approximates wheeled vehicle motion under specific control inputs on flat ground using ordinary differential equations. We use a singularity-free and differentiable variant of the single-track model to enable seamless integration as dynamics factor into VIO and to optimize the model parameters online together with the VIO state variables. We validate our method with real-world data in both indoor and outdoor environments with different terrain types and wheels. In experiments, we demonstrate that ST-VIO can not only adapt to wheel or ground changes and improve the accuracy of prediction under new control inputs, but can even improve tracking accuracy.

Online Calibration of a Single-Track Ground Vehicle Dynamics Model by Tight Fusion with Visual-Inertial Odometry

TL;DR

ST-VIO tightly fuses a differentiable, singularity-free single-track dynamics model with visual–inertial odometry to enable online online calibration of vehicle dynamics parameters while estimating VIO states. By embedding the dynamics as a multistep RK-integrated factor within a sliding-window graph and enforcing planar and geometry constraints, the method achieves improved tracking accuracy and significantly better forward-prediction under new control inputs. The approach adapts to changes in terrain and wheel properties, demonstrated across indoor/outdoor scenarios, stop-and-go maneuvers, and wheel swaps, while maintaining real-time performance. This work advances capability for model-predictive control and navigation planning in ground robots by providing online, data-driven adaptation of the vehicle dynamics model within the VIO framework.

Abstract

Wheeled mobile robots need the ability to estimate their motion and the effect of their control actions for navigation planning. In this paper, we present ST-VIO, a novel approach which tightly fuses a single-track dynamics model for wheeled ground vehicles with visual inertial odometry (VIO). Our method calibrates and adapts the dynamics model online to improve the accuracy of forward prediction conditioned on future control inputs. The single-track dynamics model approximates wheeled vehicle motion under specific control inputs on flat ground using ordinary differential equations. We use a singularity-free and differentiable variant of the single-track model to enable seamless integration as dynamics factor into VIO and to optimize the model parameters online together with the VIO state variables. We validate our method with real-world data in both indoor and outdoor environments with different terrain types and wheels. In experiments, we demonstrate that ST-VIO can not only adapt to wheel or ground changes and improve the accuracy of prediction under new control inputs, but can even improve tracking accuracy.
Paper Structure (16 sections, 4 equations, 4 figures, 2 tables)

This paper contains 16 sections, 4 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Single-Track Dynamics Model
  5. Integration of Single-Track Dynamics with VIO
  6. Geometry Constraints
  7. Optimization
  8. Offline Initial Guess Estimation
  9. EXPERIMENTS
  10. Experiment Setup
  11. Tracking Accuracy Evaluation
  12. Prediction Accuracy Evaluation
  13. Change of Robot Properties
  14. Stop-and-Go Motion
  15. Run-Time Evaluation
  16. ...and 1 more sections

Figures (4)

  • Figure 1: ST-VIO performs windowed optimization (blue box) with marginalization of old states (gray box) to estimate vehicle motion and parameters of a single-track dynamics model. The dynamics model is used as factor in the optimization through ODE integration (green box, wheels: brown rectangles, velocity: green, force: red, x-axis: longitudinal, y-axis: lateral axis).
  • Figure 2: Factor graph of ST-VIO. Blue/white circles: keyframe/recent frame variables; light green: first active recent frame at $t_0$. The dynamics factor (red) connects poses, velocities, gyroscope biases, extrinsic poses of all active recent frames and the dynamics model parameters at $t_0$.
  • Figure 3: Left: Our mobile robot is a modified 1/10 electric RC car equipped with an Intel Realsense T265 stereo camera. Right: We primarily use the bottom wheel in our experiments and also evaluate with wheels without rubber tire (top).
  • Figure 4: Top left: online calibration (calib) by ST-VIO for the new wheels clearly improves prediction over offline calibration (init) for the old wheels. Top right: evolution of online calibrated parameters (calib). Bottom left: 10 s prediction results (red: calib, blue: init, yellow trajectory: ground truth, red/purple circle: start/end, rotated by $30\degree$ for visualization). Bottom right: control inputs.