DESCRIPTION

This tutorial addresses one of the timely topics in the field of autonomous multi-vehicle systems for Smart Cities and Industry 4.0: formation control and collision avoidance for multi-agent configuration. As it is well-known, such issues have received increasing attention, as testified by the relevant number of contributions, see, e.g., [R1]–[R3] and references therein. Essentially, this is due to the success of using Unmanned Vehicles (UVs) in different areas of Industry 4.0 and Intelligent Transportation Systems, see, e.g., [R4]. In this context, it is important to underline that the design of a fully autonomous vehicle system must take into account the following key requirements: (i) each vehicle must be able to track a reference signal and reach the desired goal, (ii) formation control tools are needed to guarantee coordination among vehicles, and (iii) absence of collisions must be ensured.

Despite the significant interest, most of the existing approaches provide efficient solutions only when physical limitations are not taken into account in the vehicles’ modelling. As a consequence, their effectiveness and absence of collision claims are not guaranteed in real scenarios where the vehicle’s geometrical constraints and saturation effects play a crucial role. 

According to these premises, the tutorial is devoted to disseminating recent advances in constrained control of multi-vehicle systems. We will discuss recent solutions addressing the collision-free reference tracking control problem for constrained multi-vehicle systems moving in shared environments (e.g. Industrial Plants and/or Cities) where obstacles might be present. The distinctiveness of this tutorial relies on showing that traditional Model Predictive Control (MPC) solutions can be combined with recent-developed Reinforcement Learning tools to derive novel and more efficient hybrid control architectures for multi-vehicle autonomous systems.

TUTORIAL ORGANIZATION, TIMELINE, PRESENTATION MATERIAL

The tutorial will consist of four lectures. To encourage discussions, a Q&A period of 10/15 minutes will be available at the end of each part. Interaction between speakers and attendees will be strongly encouraged also during the talks.

• Part 1: Background material on the MPC philosophy is reviewed
  • Duration: 25 Minutes
  • Presenter: Dr. Walter Lucia
  • Slides part 1 (draft)
• Part 2: By considering a system of heterogeneous multi-unmanned vehicles moving in a shared environment, the MPC-based inter-vehicle collision avoidance strategy introduced in [R5]-[R6] is presented. Future research directions and extensions will be discussed
  • Duration: 45 Minutes
  • Speaker: Dr. Walter Lucia
  • Slides part 2 (draft)
  • Demo of the presented solution: https://youtu.be/SCeoJyle_VU
• Part 3: Background material on Neural Networkd (NN) and Reinforcement Learning is reviewed
  • Duration: 30 Minutes
  • Speaker: Dr. Giuseppe Franzè
  • Slides part 3 (draft)
• Part 4: By considering a platoon of autonomous vehicles operating in an urban road network, the distributed control architecture based on the joint use of MPC and Reinforcement Learning introduced in [R7] is presented. Future research directions and extensions are discussed
  • Duration: 45 Minutes
  • Speaker: Dr. Giuseppe Franzè
  • Slides part 4 (draft)
  • Demo of the presented solution: https://www.youtube.com/watch?v=YcxVJctk75k

RELEVANCE OF THE PROPOSED TUTORIAL

Over the past recent years, the application of autonomous vehicles in different safety-critical domains, such as defence and space, search and rescue, health care, and industry 4.0, has gained increasing interest. Therefore, the research community has made a non-trivial effort to enlarge the applicability range of AVs within networked and distributed control setups. Accordingly, big corporations like Amazon, Google, and Tesla are carrying out their in-house research and prototyping AV solutions. The results of such actions are testified by the proliferation and diffusion of autonomous vehicles in both industry and society. Very popular examples of this phenomenon are the Tesla car (autonomous vehicle) and the Amazon warehouse robots (multi-vehicle systems). Therefore, this topic appears extremely relevant for its intrinsic interdisciplinary nature in both academia and industry contexts. The tutorial will help attendees to be familiar with the latest advances in the field and to open the doors to the next generation of control strategies for multi-agent vehicle systems.

REFERENCES

[R1] K.-K. Oh, M.-C. Park, and H.-S. Ahn, “A survey of multi-agent formation control,” Automatica, 53: 424-440, 2015.

[R2] J. Dahl, G. R. de Campos, C. Olsson, J. Fredriksson, “Collision avoidance: A literature review on threat-assessment techniques’,’ IEEE Transactions on Intelligent Vehicles 4(1): 101-113, 2018.

[R3] B. Paden, M. Čáp, S. Z. Yong, D. Yershov, E. Frazzoli, “A survey of motion planning and control techniques for self-driving urban vehicles,” IEEE Transactions on intelligent vehicles 1(1):33-55, 2016.

[R4] A. Schumacher, N. Tanja, S. Wilfried Sihn, “Roadmapping towards industrial digitalization based on an Industry 4.0 maturity model for manufacturing enterprises,” Procedia Cirp 79: 409-414, 2019.

[R5] M. Bagherzadeh, S. Savehshemshaki, and W. Lucia. Guaranteed Collision-Free Reference Tracking in Constrained Multi Unmanned Vehicle Systems. IEEE Transactions on Automatic Control, 67(6):3083–3089, 2022

[R6] S. Savehshemshaki and W. Lucia. A Robust Receding-Horizon Collision Avoidance Strategy for Constrained Unmanned Ground Vehicles Moving in Shared Planar Environments. IEEE Conference on Decision and Control (CDC), 2022 (to appear)

[R7] F. Giannini, G. Fortino, G. Franzè, and F. Pupo. A Deep Q Learning-Model Predictive Control Approach to vehicle routing and control with platoon constraints. In IEEE International Conference on Automation Science and Engineering (CASE), pages 563–568. IEEE, 2022