Freitag, 30. Juni 2023, 16.30 Uhr
Synchronization-Based Cooperative Trajectory Planning of Networked Vehicles
- Maximilian Kloock, M.Sc. - Informatik 11, Embedded Software
- Ort: Raum 2202 (HBau, 2. Stock), Ahornstr. 55
Centralized decision-making for a Networked Control System (NCS) suffers from a
high computational burden on the planning agent. Distributed agents, which compute
distributed decision-making, increase the computational performance of the NCS. In
distributed decision-making, agents consider a subset of all agents in their planning.
Due to only local system knowledge of the agents, these local plans are potentially
inconsistent with the local plans of other agents. Such an inconsistency may lead to the
infeasibility of local plans. This thesis proposes an algorithm that utilizes synchronization
methods to achieve consistency of local plans in distributed decision-making. The
algorithm enables fast computations due to parallel decision-making and achieves feasible
decisions. The parallel and iterative algorithm alternates two steps: local planning and
global synchronization. In the local planning step, the agents parallelly compute local
plans. Subsequently, consistency of the local plans across agents is achieved using global
synchronization. The globally synchronized plans act as reference decisions for the local
planning step in the next iteration. In each iteration, the local planning guarantees
locally feasible plans, while the synchronization guarantees globally consistent plans.
The parallel algorithm converges to globally feasible decisions if the coupling topology
is feasible. This thesis introduces requirements for the coupling topology to achieve
convergence to globally feasible decisions and proposes a centralized and a distributed
method to generate coupling topologies that fulfill these requirements. The centralized
coupling topology generation method uses its knowledge about all agents to generate
time-invariant coupling topologies in a greedy manner. The distributed coupling topology
generation method generates time-variant coupling topologies using game-theory.
In order to demonstrate the distributed decision-making algorithm in experiments, this
thesis proposes an architecture for networked control systems that achieves deterministic
and reproducible experiments, even with non-deterministic computation and communication
times. The Cyber-Physical Mobility Lab (CPM Lab), an open-source and remotely
accessible platform for networked and autonomous vehicles, implements this architecture.
Moreover, the CPM Lab provides a suitable interface for rapid prototyping with seamless
simulations and experiments.
This thesis demonstrates the complete pipeline from uncoupled agents to globally
feasible solutions. The distributed decision-making uses a Model Predictive Control
(MPC) framework. The evaluation of this thesis uses the CPM Lab as an evaluation
platform to evaluate the Distributed MPC (DMPC) algorithm using the coupling topologies
generated from the centralized and distributed generation methods. The evaluation
shows that the proposed DMPC algorithm achieves feasible solutions while requiring
lower computation times and lower communication demands than Centralized MPC
(CMPC). The resulting trajectories show only small deviations from the globally optimal
trajectories generated by CMPC.
Es laden ein: die Dozentinnen und Dozenten der Informatik