Double-Layer Filtering Framework for High-Precision Robot Odometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current visual-inertial odometry estimation methods struggle to achieve high-precision pose estimation for mobile robots, especially in fast and violent motion scenarios, due to noise accumulation and zero drift in IMU measurements, and are resource-intensive, making them unsuitable for platforms with limited computing resources.
Innovation Solution
A double-layer filtering framework that integrates IMU, binocular camera, and legged odometer data using a Multi-state Constrained Kalman filter (MSC-KF) and Extended Kalman Filter (EKF) algorithms to correct and update state vectors, ensuring accurate and real-time odometry estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual-inertial odometry estimation methods are used, then position and attitude estimation can be achieved, but noise accumulation and zero drift in IMU measurements lead to declining accuracy in fast and violent motion scenarios
Solution Approach 1:
The patent segments the filtering process into two distinct layers: a bottom-layer filter that processes high-frequency IMU data and a top-layer filter that processes lower-frequency visual data. This segmentation allows each layer to handle data at its optimal frequency, preventing noise accumulation in the IMU integration while maintaining accurate pose estimation through the hierarchical correction structure.
Solution Approach 2:
The patent introduces a temporal dimension to the filtering architecture by processing IMU data at high frequency (bottom layer) and visual data at lower frequency (top layer). This multi-temporal-resolution approach allows the system to capture rapid motions accurately while filtering out high-frequency noise, resolving the contradiction between responsiveness and accuracy in fast motion scenarios.
2Measurement precision
If tight coupling multi-sensor fusion algorithm is used, then odometry estimation precision is improved, but algorithm complexity increases
Solution Approach 1:
The patent divides the complex tight coupling fusion algorithm into two manageable filter layers. The bottom-layer filter handles IMU data with a simpler integration approach, while the top-layer filter incorporates visual corrections. This segmentation reduces the computational complexity of each individual filter while maintaining the benefits of tight coupling fusion.
Solution Approach 2:
The patent resolves algorithmic complexity by introducing a hierarchical temporal structure. Instead of processing all sensor data simultaneously at one frequency, the system processes IMU and visual data at different temporal resolutions through two filter layers, simplifying the computational burden while achieving precise tight coupling fusion.
3Measurement precision
If iterative optimization methods are used, then pose estimation accuracy is improved, but computing resources are excessive for resource-constrained platforms
Solution Approach 1:
The patent extracts the computationally intensive iterative optimization step and replaces it with closed-form Kalman filter solutions at both filter layers. This extraction eliminates the heavy computational burden of iterative optimization while maintaining accurate pose estimation through the efficient Kalman filter mathematics.
Solution Approach 2:
The patent replaces expensive iterative optimization computations with cheaper closed-form Kalman filter calculations. The closed-form solutions provide sufficient accuracy for real-time applications on resource-constrained platforms without the excessive computational resource consumption of iterative methods.
Data Source
AI summary
A high-precision odometry estimation method based on a double-layer filtering framework, comprising the following steps: acquiring sensing information of a binocular camera, a main body IMU, a legged auxiliary IMU, a legged joint motor and a foot end force sensor; fusing the rotation angle and rotational velocity of the legged joint motor, and the measurement information of the legged auxiliary IMU and the foot end force sensor by a bottom-layer legged filter to obtain a velocity observation in a legged auxiliary IMU coordinate system; then inputting the velocity observation into an upper-layer filter, and fusing the acquired observation information of the main body IMU and binocular camera by the upper-layer filter to complete information fusion and outputting an odometry estimation result of a robot. The method can provide an effective solution for the autonomous positioning of the quadruped mobile robot in the rapid movement.