Learning-based motivation dynamics for high performance multi-robot systems
• Adnane Saoud (L2S – CentraleSupelec), firstname.lastname@example.org
• Antonio Loria (L2S – CNRS), email@example.com
Multi-robot systems are systems where robots interact among themselves to act cooperatively. They can efficiently carry out complex tasks in a variety of areas such as manufacturing, health, and intelligent transportation systems, to name a few [2, 4]. Therefore, fundamental research on the design and control of multi-robot systems constitutes a major challenge for the upcoming years, which requires strong collaborations between researchers from robotics, control theory, machine learning, and computer science.
The main approach has been to incorporate tools from single-robot control to the multi-robot set up according to the properties of the robots in hand. These properties can be described for example in terms of the robot dynamics and its actuation capabilities, as well as its communication and sensing limitations [8, 1]. The control design considers these properties while aiming at providing stabilization and tracking guarantees for the multi-robot system . However, many applications involve more complex tasks that may not be cast as a classic control objective, but rather involve a higher level of specification definition and planning. A current trend is to employ tools from computer science such as Signal Temporal Logic  (i.e., a type of formal verification language) to specify more general task specifications that induce a sequence of control actions rather than a stand-alone traditional control objective. In this context, standard tools from motion planning  and symbolic control  can be used. However, such approaches are generally based on state-space discretization which is inevitably incomplete since it is inherently rigid and depends on the models of both, the agent and its environment. Moreover, these approaches are generally suffering from scalability issues and would fail to deal with large-scale multi-robot systems. The very recent concept of motivation dynamics  may solve these difficulties in an orthgonal way. It is a continuous dynamical system that reactively composes low-level control vector fields using valuation functions designed based on value-sensitive decision-making models (e.g., ). Due to the continuous internal representation of the selection process, the motivation dynamics can be considered as a useful alternative to the existing hybrid framework, especially in situations where the control operates at a low level close to the physical hardware.
This project aims to develop a new paradigm that enhances the high performance of the future generation of multirobot systems. A novel theoretical and algorithmic framework that synthesizes high-level decision-making models with low-level motion control will be built based on cutting-edge techniques such as motivation dynamics and Signal Temporal Logic. Moreover, since it is challenging to design the parameters of the motivation dynamics, especially for complex tasks in dynamic environments, in this project, we aim at using learning-based approaches to learn these parameters , while providing formal proofs such that the learned parameters will make it possible to ensure thesatisfaction of the required specification.
Duration: 1-year postdoc at L2S
– PhD degree in Control theory, Robotics or related fields
– Experience of publishing high quality research papers
– Speaking and writing English at the scientific and professional level
– Good communication skills and ability to cooperate
Desired but not mandatory:
– Experience in multi-agent control systems
– Experience in robotic systems
Deadline for applications: June 1st, 2022
 Jorge Cortes and Magnus Egerstedt. Coordinated control of multi-robot systems: A survey. SICE Journal of Control, Measurement, and System Integration, 10(6):495–503, 2017.
 Marco Dorigo, Guy Theraulaz, and Vito Trianni. Reflections on the future of swarm robotics. Science Robotics, 5(49), 2020.
 Kazumune Hashimoto, Adnane Saoud, Masako Kishida, Toshimitsu Ushio, and Dimos Dimarogonas. Learningbased symbolic abstractions for nonlinear control systems. arXiv preprint arXiv:2004.01879, 2020.
 Mohammad R Jahanshahi, Wei-Men Shen, Tarutal Ghosh Mondal, Mohamed Abdelbarr, Sami F Masri, and Uvais A Qidwai. Reconfigurable swarm robots for structural health monitoring: a brief review. International Journal of Intelligent Robotics and Applications, 1(3):287–305, 2017.
 Steven M LaValle. Planning algorithms. Cambridge university press, 2006.
 Antonio Loria, Janset Dasdemir, and Nohemi Alvarez Jarquin. Leader–follower formation and tracking control of mobile robots along straight paths. IEEE transactions on control systems technology, 24(2):727–732, 2015.
 Oded Maler and Dejan Nickovic. Monitoring temporal properties of continuous signals. In Formal Techniques, Modelling and Analysis of Timed and Fault-Tolerant Systems, pages 152–166. Springer, 2004.
 Emmanuel Nu˜no, Antonio Lor´ıa, Elena Panteley, and Esteban Restrepo. Rendezvous of nonholonomic robots via output-feedback control under time-varying delays. IEEE Transactions on Control Systems Technology, 2022.
 Darren Pais, Patrick M Hogan, Thomas Schlegel, Nigel R Franks, Naomi E Leonard, and James AR Marshall. A mechanism for value-sensitive decision-making. PloS one, 8(9):e73216, 2013.
 Paul Reverdy and Daniel E Koditschek. A dynamical system for prioritizing and coordinating motivations. SIAM Journal on Applied Dynamical Systems, 17(2):1683–1715, 2018.
 Adnane Saoud. Compositional and efficient controller synthesis for cyber-physical systems. PhD thesis, Universite Paris-Saclay (ComUE), 2019.