Mobile Robot Pose Estimation Using Particle Filter and Odometry

The patent changes the fundamental parameters used for pose estimation from feature-point-based methods to odometry-based methods with uncertainty modeling. By switching to wheel encoder data and incorporating covariance matrices to represent uncertainty, the system achieves robust performance in environments where traditional feature-point methods fail, directly addressing the contradiction between accuracy and efficiency across different environments.

The patent introduces dynamic uncertainty modeling where the covariance matrix is continuously updated based on odometry data quality and environmental conditions. This dynamic adaptation allows the system to adjust its confidence in pose estimates in real-time, maintaining efficiency while achieving acceptable accuracy even in feature-poor environments by relying more on odometry when features are unavailable.

The patent prepares for potential odometry errors by incorporating uncertainty covariance matrices in advance. The system models expected error accumulation on different surface types (slippery floors, carpets) and prepares appropriate covariance values before errors occur. This allows the filtering algorithm to weigh odometry data appropriately and compensate for expected drift, cushioning against reliability degradation while maintaining real-time performance.

The patent implements a feedback mechanism where the system continuously monitors the consistency between odometry-based pose estimates and actual sensor observations. When discrepancies indicate error accumulation (such as unexpected deviations from expected feature relationships), the system adjusts its confidence in odometry data and increases reliance on sensor feedback, thereby correcting drift while maintaining real-time localization capability.

The patent extracts and removes the computationally intensive feature extraction and matching step from the pose estimation process. By eliminating this step entirely and relying on odometry with uncertainty modeling, the system dramatically reduces memory and computation requirements while maintaining acceptable pose estimation accuracy, directly resolving the contradiction between precision and device complexity.

The patent replaces expensive, computation-intensive feature matching with cheaper, simpler odometry calculations. The system uses readily available wheel encoder data that requires minimal processing, effectively substituting a low-cost, short-living approximation (odometry) for the expensive, resource-intensive feature matching approach, thereby reducing device complexity while maintaining functional adequacy.