Research on Improved Gaussian Smoothing Filters for SLAM Application

To address the navigation issues of the planetary rover and construct a map for the unknown environment as well as the surface of the planets in our solar system, the simultaneous localization and mapping can be seen as an alternative method. In terms of the navigation with the laser sensor, the Kalman filter and its improving algorithms, such as EKF and UKF are widely used in the the process of processing information. Nevertheless, these filter algorithms suffer from low accuracy and significant computation expensive. The EKF algorithm has a linearization process, the UKF algorithm is better matched in a nonlinear system than the EKF algorithm, but it has more computational complexity. The GP-RTSS filtering algorithm, based on a Gaussian filter, is significantly superior to the EKF and UKF algorithms regarding the sensor fusion accuracy. The Gaussian Process can be used in different non-linear system, does not need prediction model and linearization. However, the main barrier in the process of implementing the GP-RTSS algorithm is that the Gaussian core function requires a lot of computation. In this paper, an algorithm, so-called DIS RTSS filter under a distributed computation scheme, derived from the GP-RTSS Gaussia n smoothing and filter, is proposed. The distributed system can effectively reduce the cost of computation (computation expense and memory). Moreover, four fusion methods for the DIS RTSS filter, i.e., DIS RTP, DIS RTGP, DIS RTB, DIS RTrB are discussed in this paper. The experiments show that among the four algorithms described above, the DIS RTGP algorithm is the most effective solution for practical implementation. The DIS RTSS filtering algorithm can realize a high processing rate and can theoretically process an infinite number of data samples.

1. Durrant W. H., Bailey T. Simultaneous localization and mapping, IEEE robotics & automation magazine, 2006, vol. 13, pp. 99—110.

2. Barfoot T. D. State Estimation for Robotics, Cambridge University Press, 2017, pp. 1—368.

3. Ellery A. Planetary rovers: robotic exploration of the solar system, Springer, 2016, pp. 1—702.

4. Achin J., Truong X. N., Morari M. Learning and control using Gaussian processes, Submitted to ACM/IEEE 9th International Conference on Cyber-Physical Systems (ICCPS). IEEE, 2018, pp. 140—149.

5. Deisenroth M. P., Rasmussen C. E., Fox D. Learning to control a low-cost manipulator using data-efficient reinforcement learning, Science & Systems, 2011.

6. Kocijan J. Modelling and control of dynamic systems using Gaussian process models, Springer, 2011, pp. 57—64.

7. Cao G., Lai E. M. K., Alam F. Gaussian process model predictive control of unknown non-linear systems, IET Control Theory & Applications, 2017, vol. 11(5), pp. 703—713.

8. Deisenroth M. P., Turner R. D., Huber M. F. et al. Robust filtering and smoothing with Gaussian processes, IEEE Transactions on Automatic Control, 2012, 57(7), pp. 1865—1871.

9. Arasaratnam I., Haykin S. Cubature Kalman Filters, IEEE Transactions on Automatic Control, 2009, vol . 54, no. 6, pp. 1254—1269.

10. Quinonero-Candela J., Rasmussen C. E. A unifying view of sparse approximate Gaussian process regression, Journal of Machine Learning Research, 2005, vol. 6(12), pp. 1939—1960.

11. Fomichev A. V., Wang Guoyan. Jeffektivnyj algoritm filhtra Gaussa dlja metoda SLAM, 11 Rossijskaja mulhtikonferenciija po problemam upravlenija "Upravlenie v ajerokosmichskih sistemah" (UAS-2018), 2018, pp. 260—269 (in Russian).

12. Nguyen Trung V., Bonilla E. V. Fast allocation of Gaussian process experts, Proceedings of the International Conference on Machine Learning, 2014, pp. 145—153.

13. Cao Y., Fleet D. J. Generalized product of experts for automatic and principled fusion of Gaussian process predictions, preprint arXiv:1410.7827, 2014.

14. Tresp V. A Bayesian сommittee machine, Neural Computation, 2000.

15. Deisenroth P. J., Ng W. Distributed Gaussian processes, Preprint arXiv:1502.02843, 2015.