IEEE Robotics & Automation Magazine - September 2023 - 55

while minimizing oscillations. Furthermore, the neural network
is fed explicitly with both measured and observed variables, making
the learning process easier.
Both methods have been tested through full-scale experiments
using a mobile robot acting in off-road conditions at
relatively high speeds on different kinds of terrain. The experimental
results demonstrate that both strategies for gain adaptation
are able to significantly outperform the result obtained
when using constant gains for the control law. Additionally,
experiments show that the neural adaptation approach is also
capable of outperforming the deterministic adaptation gain
method in the described trajectories. This result is made possible
with the addition of estimated grip conditions to the network's
inputs (the cornering stiffness observer). The use of
virtual batch normalization as well as an adaptive linear tyre
model, rather than a more complex nonlinear model (such as
Pacejka), also constitute a contribution, which produces a satisfactory
and repeatable result.
The neural gain method was trained without any use of
prior expert knowledge, and from this it is able to determine
online and in real time the control parameters that minimize an
obj function in a highly dynamic environment. Unfortunately,
the neural gain method does have some significant drawbacks,
such as training time and the black-box nature of neural networks,
preventing a thorough understanding of the procedure's
behavior. This implies that a neural network cannot be explicitly
proven to be stable in all conditions and over every trajectory.
Indeed, should the system delay vary between the training
simulation and the real world, a significant loss in tracking
accuracy can be expected. Moreover, the method is limited
due to training with a constant longitudinal speed, and as such,
strong variations of longitudinal speed can cause instabilities.
Accordingly, future works are focused on modulation of
the target speed, together with the proposed approach for
adapting control parameters. The objective is to preserve
the accuracy of trajectory tracking under dynamic and
uncertain conditions, using a neural network adaptation of
velocity and steering control parameters. This enables the
use of stable, explainable, and deterministic control laws
while contributing to the field of artificial intelligence
applied to robotics.
ACKNOWLEDGMENTS
This article was made possible by the use of Factory-IA cluster
and financially supported by the Ile-de-France Regional
Council. It received support from the French government
research program " Investissements d'Avenir " through the Initiative
d'Exellence (IDEX)-Initiatives-Science-Innovation-
Territories- Economy initiative 16-IDEX-0001 (CAP 20-25)
and the Innovative Mobility: Smart and Sustainable Solutions
Laboratory of Excellence (ANR-10-LABX-16-01).
AUTHORS
Ashley Hill, Université Paris-Saclay, French Alternative
Energies and Atomic Energy Commission, Palaiseau, 91120,
France. E-mail: ashley.hill.1993@cea.fr.
Jean Laneurit, Université Clermont Auvergne, Inrae,
Unité de recherche Technologies et systèmes d'information
pour les agro-systèmes, Centre de Clermont-Ferrand,
Aubière, 63178, France. E-mail: jean.laneurit@irstea.fr.
Roland Lenain, Université Clermont Auvergne, Inrae,
Unité de recherche Technologies et systèmes d'information
pour les agro-systèmes, Centre de Clermont-Ferrand,
Aubière, 63178, France. E-mail: roland.lenain@inrae.fr.
Eric Lucet, Université Paris-Saclay, French Alternative
Energies and Atomic Energy Commission, Palaiseau, 91120,
France. E-mail: eric.lucet@cea.fr.
REFERENCES
[1] F.-J. Lin and C.-H. Lin, " On-line gain-tuning IP controller using RFNN, "
IEEE Trans. Aerosp. Electron. Syst., vol. 37, no. 2, pp. 655-670, 2001, doi:
10.1109/7.937476.
[2] M. Tursini, F. Parasiliti, and D. Zhang, " Real-time gain tuning of PI controllers
for high-performance PMSM drives, " IEEE Trans. Ind. Appl., vol. 38, no. 4,
pp. 1018-1026, 2002, doi: 10.1109/TIA.2002.800564.
[3] N. T. Binh, N. A. Tung, D. P. Nam, and N. H. Quang, " An adaptive backstepping
trajectory tracking control of a tractor trailer wheeled mobile robot, " Int. J. Contr.,
Automat. Syst., vol. 17, no. 2, pp. 465-473, 2019, doi: 10.1007/s12555-017-0711-0.
[4] P. Panahandeh, K. Alipour, B. Tarvirdizadeh, and A. Hadi, " A self-tuning trajectory
tracking controller for wheeled mobile robots, " Ind. Robot, Int. J. Robot.
Res. Appl., vol. 46, no. 6, pp. 828-838, 2019, doi: 10.1108/IR-02-2019-0032.
[5] A. Hill, J. Laneurit, R. Lenain, and E. Lucet, " Online gain setting method for
path tracking using CMA-ES: Application to off-road mobile robot control, " in
Proc. 2020 IEEE/RSJ Int. Conf. Intel. Robots Syst. (IROS), pp. 7697-7702, doi:
10.1109/IROS45743.2020.9340830.
[6] D. Chindamo, B. Lenzo, and M. Gadola, " On the vehicle sideslip angle estimation:
A literature review of methods, models, and innovations, " Appl. Sci., vol. 8,
no. 3, p. 355, 2018, doi: 10.3390/app8030355.
[7] R. Lenain, B. Thuilot, C. Cariou, and P. Martinet, " Adaptive and predictive
path tracking control for off-road mobile robots, " Eur. J. Control, vol. 13, no. 4,
pp. 419-439, 2007, doi: 10.3166/ejc.13.419-439.
[8] M. Deremetz, R. Lenain, B. Thuilot, and V. Rousseau, " Adaptive trajectory
control of off-road mobile robots: A multi-model observer approach, " in Proc.
IEEE Int. Conf. Robot. Automat. (ICRA), May 2017, pp. 4407-4413, doi: 10.1109/
ICRA.2017.7989509.
[9] L. Jetto, S. Longhi, and D. Vitali, " Localization of a wheeled mobile robot by
sensor data fusion based on a fuzzy logic adapted Kalman filter, " Control Eng.
Pract., vol. 7, no. 6, pp. 763-771, 1999, doi: 10.1016/S0967-0661(99)00028-3.
[10] R. N. Jazar, Vehicle Dynamics: Theory and Application. New York, NY,
USA: Springer Science & Business Media, 2017.
[11] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, MA,
USA: MIT Press, 2016.
[12] T. Salimans, J. Ho, X. Chen, S. Sidor, and I. Sutskever, " Evolution strategies
as a scalable alternative to reinforcement learning, " 2017, arXiv:1703.03864.
[13] R. Lenain, M. Deremetz, J.-B. Braconnier, B. Thuilot, and V. Rousseau,
" Robust sideslip angles observer for accurate off-road path tracking control, " Adv.
Robot., vol. 31, no. 9, pp. 453-467, 2017, doi: 10.1080/01691864.2017.1280414.
[14] E. Bakker, L. Nyborg, and H. B. Pacejka, " TYRE modelling for use in vehicle
dynamics studies, " SAE International, Warrendale, PA, USA, Tech. Paper
870421, 1987.
[15] R. S. Sutton and A. G. Barto, Introduction to Reinforcement Learning, vol. 2.
Cambridge, MA, USA: MIT Press, 1998.
[16] R. S. Sutton, " Learning to predict by the methods of temporal differences, "
Mach. Learn., vol. 3, no. 1, pp. 9-44, 1988, doi: 10.1007/BF00115009.
[17] N. Hansen, " The CMA evolution strategy: A tutorial, " 2016, arXiv:1604.00772.
[18] X. Zhang, J. Clune, and K. O. Stanley, " On the relationship between the
OpenAI evolution strategy and stochastic gradient descent, " 2017, arXiv:
1712.06564.
[19] A. Hill, E. Lucet, and R. Lenain, " A new neural network feature importance
method: Application to mobile robots controllers gain tuning, " in Proc. 2020 17th
Int. Conf. Inf. Control, Automat. Robot., ICINCO, pp. 188-194, doi:
10.5220/0009888501880194.
SEPTEMBER 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE
55
IEEE Robotics & Automation Magazine - September 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2023
