Periodic Reporting for period 2 - LEAFHOUND (Leader-follower hybrid control and task planning for multi-agent systems under spatiotemporal logic specifications)
Okres sprawozdawczy: 2022-05-01 do 2023-10-31
Current applications of multi-agent systems involve more complex tasks that may not be cast as classic control objectives, but rather involve a higher level of specification definition and planning. The robots need to fulfill high level tasks and also taking into account their different capabilities. Hence, both task planning for high level specifications and continuous control are addressed in this project. Temporal logics, such as Linear Temporal Logic (LTL) and Signal Temporal Logic (STL), provide a formalism to mathematically formulate such tasks. We consider in this project a heterogeneous, leader-follower setting, in which selected agents with advanced actuation, computation and communication capabilities, namely the leaders, are responsible for guiding the whole group to implement the tasks in a decentralized way while fulfilling the transient constraints.
The research of (leader-follower) multi-agent systems has various important applications including search and rescue missions, collaborative transportation, assistance for healthcare, as well as intelligent transportation networks. In terms of leader-follower framework, it shows advantages such as scalability, adaptability to changes in workspace, fault tolerance, reduction of computation or communication cost, reliability and robustness. It can also help to allocate tasks and resources more reasonably. Besides, the high level spatiotemporal specification which considers both space and time constraints is crucial to handle certain safety-critical applications.
The general objective is to propose a novel approach to leader-follower multi-agent control under spatiotemporal logic specifications. Along this line, the project's objectives are summarized in (O1) leader-follower transient control of coupled multi-agent systems, (O2) task decomposition and planning for leader-follower multi-agent systems under spatiotemporal logic specifications, (O3) hybrid control of leader-follower multi-agent systems under dependent or infeasible spatiotemporal logic specifications, (O4) graph theoretic controllability approach to leader-follower multi-agent systems, (O5) application to leader-follower multi-agent systems which involves validating and demonstrating the overall scheme.
The research of (leader-follower) multi-agent systems has various important applications including search and rescue missions, collaborative transportation, assistance for healthcare, as well as intelligent transportation networks. In terms of leader-follower framework, it shows advantages such as scalability, adaptability to changes in workspace, fault tolerance, reduction of computation or communication cost, reliability and robustness. It can also help to allocate tasks and resources more reasonably. Besides, the high level spatiotemporal specification which considers both space and time constraints is crucial to handle certain safety-critical applications.
The general objective is to propose a novel approach to leader-follower multi-agent control under spatiotemporal logic specifications. Along this line, the project's objectives are summarized in (O1) leader-follower transient control of coupled multi-agent systems, (O2) task decomposition and planning for leader-follower multi-agent systems under spatiotemporal logic specifications, (O3) hybrid control of leader-follower multi-agent systems under dependent or infeasible spatiotemporal logic specifications, (O4) graph theoretic controllability approach to leader-follower multi-agent systems, (O5) application to leader-follower multi-agent systems which involves validating and demonstrating the overall scheme.
In the reporting period, we had substantial progress on O1. We employ control barrier functions (CBF) for ensuring the satisfaction of transient and spatiotemporal constraints. In particular, we propose a method for their distributed implementation for an STL fragment, we consider their application to leader-follower networks and propose nonsmooth CBFs for decision making. Followed by we devised a distributed control scheme to effectively manage multiple constraints. Besides, we developed a novel PPC scheme for formation control with scaling and orientation adjustment, and a PPC scheme for leader-follower networks subject to a class of STL constraints. Moreover, we proposed model predictive control (MPC) to account for STL and coupled state constraints. Notably, we introduced corridor MPC, which was specifically designed to inspect satellite structures.
Building up on the results from O1, we derived results on task decomposition and planning for leader-follower (O2). We proposed for the decomposition of a global STL formula for a leader-follower network a convex optimization problem to compute local subtasks. We also developed a sampling-based algorithm for the generation of trajectories satisfying an STL task that couples several cooperating agents. Moreover, we proposed solutions to the problem of leader-follower multi-agent task scheduling under time constraints. We also proposed a decentralized control approach to facilitate implicit coordination among vehicles. By employing distributed model predictive control (MPC) for STL tasks, we provided a practical approach to overcome the exponential complexity associated with such tasks. Moreover, we established a method for converting STL formulas into assume-guarantee contracts, thereby enabling the satisfaction of global requirements through local contracts.
For O3, we obtained results on the hybrid control of leader-follower multi-agent systems (MAS). We propose a control scheme to robustly guarantee the fixed-time satisfaction of dependent STL tasks for coupled MAS in a least violating way in the presence of undesired violation effects of neighboring agents. Particularly, the robust performance of the task satisfactions can be adjusted in a user-specified way. Furthermore, we successfully synthesized controllers for nested STL tasks, accommodating general nonlinear systems. Additionally, we introduced a funnel-based control methodology specifically designed to address conflicting hard and soft STL constraints.
Towards O4, we derived necessary and sufficient conditions on the leader-follower graph topology under which the target formation together with the prescribed performance guarantees can be fulfilled.
For O5, we demonstrate the effectiveness and practical relevance of our developed methods by applying them to platooning (intelligent transportation) and collaborative manipulation. For the platooning application, we expressed the splitting, merging and distance maintenance tasks in one STL-task and designed a feedback controller using CBFs for systems with limited actuation capabilities. Additionally, we introduced methods for coordinating multiple platoons and robustly managing platoons under STL constraints. Moreover, we designed a low-level controller for a manipulator system handling an object under given STL specifications and unknown dynamics, which we experimentally verified using two robotic arms handling an object. We also investigated human-robot collaboration where a human takes the role of the leader agent and the robot the role of the follower agent.
Building up on the results from O1, we derived results on task decomposition and planning for leader-follower (O2). We proposed for the decomposition of a global STL formula for a leader-follower network a convex optimization problem to compute local subtasks. We also developed a sampling-based algorithm for the generation of trajectories satisfying an STL task that couples several cooperating agents. Moreover, we proposed solutions to the problem of leader-follower multi-agent task scheduling under time constraints. We also proposed a decentralized control approach to facilitate implicit coordination among vehicles. By employing distributed model predictive control (MPC) for STL tasks, we provided a practical approach to overcome the exponential complexity associated with such tasks. Moreover, we established a method for converting STL formulas into assume-guarantee contracts, thereby enabling the satisfaction of global requirements through local contracts.
For O3, we obtained results on the hybrid control of leader-follower multi-agent systems (MAS). We propose a control scheme to robustly guarantee the fixed-time satisfaction of dependent STL tasks for coupled MAS in a least violating way in the presence of undesired violation effects of neighboring agents. Particularly, the robust performance of the task satisfactions can be adjusted in a user-specified way. Furthermore, we successfully synthesized controllers for nested STL tasks, accommodating general nonlinear systems. Additionally, we introduced a funnel-based control methodology specifically designed to address conflicting hard and soft STL constraints.
Towards O4, we derived necessary and sufficient conditions on the leader-follower graph topology under which the target formation together with the prescribed performance guarantees can be fulfilled.
For O5, we demonstrate the effectiveness and practical relevance of our developed methods by applying them to platooning (intelligent transportation) and collaborative manipulation. For the platooning application, we expressed the splitting, merging and distance maintenance tasks in one STL-task and designed a feedback controller using CBFs for systems with limited actuation capabilities. Additionally, we introduced methods for coordinating multiple platoons and robustly managing platoons under STL constraints. Moreover, we designed a low-level controller for a manipulator system handling an object under given STL specifications and unknown dynamics, which we experimentally verified using two robotic arms handling an object. We also investigated human-robot collaboration where a human takes the role of the leader agent and the robot the role of the follower agent.
The obtained results so far contain several novelties, which were required to address O1-O5. In particular, the outcome of the results includes a successful merging of techniques from robust and transient control, formal verification, and multi-agent network analysis.
Regarding the results developed in O1, the main novelty relies on having controlled leaders respecting certain STL constraints for the multi-agent setup, while the followers are guided indirectly through their dynamic couplings with the leaders.Notably a variant of model predictive control (MPC) known as corridor MPC was employed for the novel task of satellite tracking. Furthermore, nonsmooth control barrier functions (CBFs) were utilized to incorporate disjunction constraints into the system.
Imposing high-level tasks to robotic systems and task decomposition (O2), has been approached previously only using formal verification languages (e.g. LTL) with discrete dynamic representations mainly for single-agent cases. In contrast, we considered the usage of continuous dynamics and STL constraints for multi-agent systems. Moreover, having a leader-follower structure, along with the employment of discretization-free methods, paves the way to a scalable solution for multi-agent task planning problem. Traditional optimization-based approaches for enforcing STL constraints suffer from exponential complexity. However, we address this issue by introducing distributed model predictive control (MPC) as a means to navigate the exponential complexity challenge. Furthermore, we present a novel method for incorporating STL constraints into assume-guarantee contracts, enhancing the overall system's flexibility and robustness.
O3 involves the treatment of potential conflicts and dependencies in the local STL tasks of the agent subgroups. No previous approach has considered dynamics and couplings between the agents and they also involved LTL and not STL specifications. Related to O3 we have obtained some preliminary results to improve the funnel control method for the single-agent case under conflicting spatiotemporal specifications. Furthermore, we have devised means to tackle nested STL tasks. These new findings serve as a significant advancement, as they pave the way for addressing the challenges posed by O3 in the multi-agent case. It is worth noting that previous research primarily focused on linear systems, while our results extend to encompass general nonlinear systems.
O4 concerns analyzing conditions of multi-agent system controllability and leader selection for leader-follower multi-agent networks under temporal logic specifications. None of the previous results on leader selection or network controllability involves LTL or STL for multi-agent systems. We have developed some preliminary results that derive topological conditions for leader-follower networks under transient constraints. Further works on O4 are left for the future.
Finally, O5 includes the applications and experiments of the developed results. Novel results are proposed for multi-robot cooperative manipulation and multi-vehicle platooning under STL constraints. We highlight that previous similar applications in the literature did not consider STL. Furthermore, we are currently developing a multi-agent manipulation algorithm, which will undergo experimental validation. This ongoing work sets the stage for future results in leader-follower systems.
Regarding the results developed in O1, the main novelty relies on having controlled leaders respecting certain STL constraints for the multi-agent setup, while the followers are guided indirectly through their dynamic couplings with the leaders.Notably a variant of model predictive control (MPC) known as corridor MPC was employed for the novel task of satellite tracking. Furthermore, nonsmooth control barrier functions (CBFs) were utilized to incorporate disjunction constraints into the system.
Imposing high-level tasks to robotic systems and task decomposition (O2), has been approached previously only using formal verification languages (e.g. LTL) with discrete dynamic representations mainly for single-agent cases. In contrast, we considered the usage of continuous dynamics and STL constraints for multi-agent systems. Moreover, having a leader-follower structure, along with the employment of discretization-free methods, paves the way to a scalable solution for multi-agent task planning problem. Traditional optimization-based approaches for enforcing STL constraints suffer from exponential complexity. However, we address this issue by introducing distributed model predictive control (MPC) as a means to navigate the exponential complexity challenge. Furthermore, we present a novel method for incorporating STL constraints into assume-guarantee contracts, enhancing the overall system's flexibility and robustness.
O3 involves the treatment of potential conflicts and dependencies in the local STL tasks of the agent subgroups. No previous approach has considered dynamics and couplings between the agents and they also involved LTL and not STL specifications. Related to O3 we have obtained some preliminary results to improve the funnel control method for the single-agent case under conflicting spatiotemporal specifications. Furthermore, we have devised means to tackle nested STL tasks. These new findings serve as a significant advancement, as they pave the way for addressing the challenges posed by O3 in the multi-agent case. It is worth noting that previous research primarily focused on linear systems, while our results extend to encompass general nonlinear systems.
O4 concerns analyzing conditions of multi-agent system controllability and leader selection for leader-follower multi-agent networks under temporal logic specifications. None of the previous results on leader selection or network controllability involves LTL or STL for multi-agent systems. We have developed some preliminary results that derive topological conditions for leader-follower networks under transient constraints. Further works on O4 are left for the future.
Finally, O5 includes the applications and experiments of the developed results. Novel results are proposed for multi-robot cooperative manipulation and multi-vehicle platooning under STL constraints. We highlight that previous similar applications in the literature did not consider STL. Furthermore, we are currently developing a multi-agent manipulation algorithm, which will undergo experimental validation. This ongoing work sets the stage for future results in leader-follower systems.