[Paper Review] Visual-Inertial Localization for Skid-Steering Robots with Kinematic Constraints

Xingxing Zuo, Mingming Zhang+4 more|arXiv (Cornell University)|Nov 13, 2019

Robotics and Sensor-Based Localization4 citations

TL;DR

This paper proposes a tightly-coupled visual-inertial localization system for skid-steering robots that explicitly models time-varying kinematic constraints using instantaneous center of rotation (ICR) dynamics and estimates them online alongside navigation states. By integrating ICR-based kinematic parameters into a sliding-window bundle adjustment framework, the method achieves significantly improved 3D localization accuracy—reducing errors by up to an order of magnitude in outdoor and indoor scenarios—especially when GPS is unavailable or degraded.

ABSTRACT

While visual localization or SLAM has witnessed great progress in past decades, when deploying it on a mobile robot in practice, few works have explicitly considered the kinematic (or dynamic) constraints of the real robotic system when designing state estimators. To promote the practical deployment of current state-of-the-art visual-inertial localization algorithms, in this work we propose a low-cost kinematics-constrained localization system particularly for a skid-steering mobile robot. In particular, we derive in a principle way the robot's kinematic constraints based on the instantaneous centers of rotation (ICR) model and integrate them in a tightly-coupled manner into the sliding-window bundle adjustment (BA)-based visual-inertial estimator. Because the ICR model parameters are time-varying due to, for example, track-to-terrain interaction and terrain roughness, we estimate these kinematic parameters online along with the navigation state. To this end, we perform in-depth the observability analysis and identify motion conditions under which the state/parameter estimation is viable. The proposed kinematics-constrained visual-inertial localization system has been validated extensively in different terrain scenarios.

Motivation & Objective

• Address the lack of explicit kinematic constraint modeling in existing visual-inertial SLAM systems for real robotic platforms.
• Improve localization accuracy for skid-steering robots by modeling time-varying ICR-based kinematics due to terrain interaction and mechanical imperfections.
• Enable robust, real-time 3D localization using low-cost sensors (camera, IMU, wheel encoders) in GPS-denied environments.
• Ensure observability of kinematic parameters under general motion through theoretical analysis.
• Validate the system across diverse terrain types, including indoor and outdoor scenarios with and without GPS.

Proposed method

• Model the kinematics of skid-steering robots using instantaneous centers of rotation (ICR) for the robot body, left track, and right track, capturing nonholonomic constraints.
• Formulate time-varying kinematic parameters (e.g., ICR positions, scale factors) to account for slippage and terrain roughness.
• Integrate the ICR-based kinematic constraints into a tightly-coupled sliding-window bundle adjustment (BA) estimator for joint optimization of visual, inertial, and odometric measurements.
• Estimate the kinematic parameters online in real time using the BA framework, treating them as state variables alongside pose and velocity.
• Perform observability analysis to prove that kinematic parameters are locally observable under general motion, especially when IMU data is available.
• Use a dual-robot setup (Clearpath Jackal) with RTK-GPS for ground truth and evaluate performance across 20 diverse sequences.

Experimental results

Research questions

• RQ1 Can online estimation of time-varying ICR-based kinematic parameters improve visual-inertial localization accuracy for skid-steering robots?
• RQ2 Under what motion conditions are the kinematic parameters of a skid-steering robot observable in a visual-inertial estimation framework?
• RQ3 How does the proposed method compare to baseline approaches that ignore or approximate kinematic constraints in real-world, GPS-denied environments?
• RQ4 To what extent does the inclusion of ICR modeling reduce drift in long-term localization?
• RQ5 Can the system maintain accurate and consistent kinematic parameter estimation under varying terrain conditions and initial parameter uncertainty?

Key findings

• The proposed method reduces final drift by up to an order of magnitude in 20 representative sequences, with significant improvements in both indoor and outdoor environments.
• In GPS-available sequences, the root mean square error (RMSE) was reduced from 9.41m (without kinematic modeling) to 0.82m (with online estimation of ξ) on sequence CP01-2019-04-19-15-42-40.
• Kinematic parameters converge rapidly to correct values even from poor initial estimates, with uncertainty envelopes shrinking quickly over time.
• The observability analysis confirms that kinematic parameters are locally observable under general motion, but observability breaks down when IMU data is excluded.
• Trajectory estimates with online ICR parameter estimation closely track RTK-GPS ground truth, as shown in visual comparisons (Fig. 5), especially in challenging sequences with complex terrain.
• The method demonstrates robustness across diverse terrains, including flat, bumpy, and uneven surfaces, validating its practical deployability.