- AI-based multi-object relative state estimation for mobile robots
- Uses DNNs to extract object class and 6-DoF relative pose from images
- Fuses measurements in EKF with uncertainty quantification and outlier rejection
- Reformulates measurement equation to decouple position and rotation
- Replaces fixed covariance with predicted aleatoric uncertainty
2. 先行研究と比べてどこがすごい？
- Prior methods use fixed measurement covariance for 6-DoF pose
- New approach decouples position and rotation, limiting erroneous rotation influence
- Enables partial measurement rejection
- Improves performance and consistency of state estimators
- Leverages DNN's predicted aleatoric uncertainty instead of fixed covariance
3. 技術・手法の肝は？
- Reformulate measurement equation for direct object-relative pose measurement
- Derive EKF to decouple position and rotation measurements
- Use predicted aleatoric uncertainty from DNN as measurement covariance
- Implement outlier rejection for partial measurements
- Integrate AI-based measurements with EKF framework
4. どうやって有効だと検証した？
- Investigate performance and consistency improvements
- Compare with fixed measurement covariance matrix
- Evaluate state estimator performance
- Assess outlier rejection capabilities
- Details of experimental setup not specified in abstract
5. 議論はある？
- Benefits of reformulating measurement equation
- Decoupling position and rotation reduces influence of erroneous rotation
- Aleatoric uncertainty improves state estimation
- Challenges in quantifying DNN uncertainty and outlier rejection
- Limitations and potential improvements not discussed in abstract
6. 次に読むべき論文は？
- Extended Kalman Filter (EKF) for state estimation
- Deep Neural Networks (DNNs) for pose estimation
- Aleatoric uncertainty quantification in DNNs
- Outlier rejection techniques in sensor fusion
- Multi-object relative state estimation methods
Precise localization with respect to a set of objects of interest enables mobile robots to perform various tasks. With the rise of edge devices capable of deploying deep neural networks (DNNs) for real-time inference, it stands to reason to use artificial intelligence (AI) for the extraction of object-specific, semantic information from raw image data, such as the object class and the relative six degrees of freedom (6-DoF) pose. However, fusing such AI-based measurements in an Extended Kalman Filter (EKF) requires quantifying the DNNs' uncertainty and outlier rejection capabilities.
This paper presents the benefits of reformulating the measurement equation in AI-based, object-relative state estimation. By deriving an EKF using the direct object-relative pose measurement, we can decouple the position and rotation measurements, thus limiting the influence of erroneous rotation measurements and allowing partial measurement rejection. Furthermore, we investigate the performance and consistency improvements for state estimators provided by replacing the fixed measurement covariance matrix of the 6-DoF object-relative pose measurements with the predicted aleatoric uncertainty of the DNN.
