Robot Orientation Estimation Using Multi-Hypothesis Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous robots face challenges in determining their orientation accurately, especially when equipped with accelerometers and gyroscopes, as they struggle to relate internal and external coordinate frames, leading to potential crashes or loss of control due to incorrect initial assumptions.
Innovation Solution
The system fuses measurements from internal and external sensing methods using multiple hypotheses, employing state estimators like Kalman Filters to compute the likelihood of each hypothesis and select the most probable orientation, thereby reducing reliance on noisy sensors like magnetometers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If magnetometers are used to determine orientation, then orientation information can be obtained, but the system becomes susceptible to external disturbances and noise
Solution Approach 1:
The patent introduces an intermediary computational process that fuses accelerometer and gyroscope measurements to estimate orientation. This intermediary estimation mechanism mediates between the unreliable magnetometer readings and the control system, filtering out external disturbances while preserving orientation information.
Solution Approach 2:
The patent replaces the direct magnetic sensing approach with an inertial-based estimation system. By substituting the magnetometer's direct measurement function with a computational model using accelerometers and gyroscopes, the system achieves orientation determination without susceptibility to magnetic disturbances.
2Measurement precision
If multiple hypotheses are evaluated using state estimators, then orientation accuracy is improved, but computational resources are consumed
Solution Approach 1:
The patent applies partial action by evaluating multiple hypotheses only to the extent necessary for accurate orientation determination. The system generates several orientation hypotheses from sensor measurements and evaluates them using state estimators, but stops once sufficient accuracy is achieved, avoiding unnecessary computational expenditure.
Solution Approach 2:
The patent changes the parameter of hypothesis evaluation by implementing a threshold-based stopping criterion. When the likelihood difference between the best hypothesis and other hypotheses exceeds a predefined threshold, the system terminates further evaluation, dynamically adjusting computational effort based on the confidence level achieved.
3Ease of manufacture
If the robot relies on initial orientation assumptions, then the system is simple to implement, but incorrect assumptions can cause crashes or loss of control
Solution Approach 1:
The patent applies preliminary action by pre-computing multiple orientation hypotheses based on initial sensor measurements before the robot begins operation. This preliminary computation establishes a set of candidate orientations that are then continuously evaluated, allowing the system to start with simple assumptions while having prepared alternative orientations for immediate selection if the initial assumption proves incorrect.
Data Source
AI summary
The system receives position information in both an internal coordinate frame and an external coordinate frame. The system uses a comparison of position information in these flames to determine orientation information. The system determines one or more orientation hypotheses, and analyzes the position information based on these hypotheses. The system may include on-board accelerometers, gyroscopes, or both that provide the measurements in the internal coordinate frame. These measurements may be integrated otherwise processed to determine position, velocity, or both. Measurements in the external frame are provided by GPS sensors or other positioning systems. Position information is transformed to a common coordinate frame, and an error metric is determined. Based on the error metric, the system estimates a likelihood metric for each hypothesis, and determines a resulting hypothesis based on the maximum likelihood or a combination of likelihoods.