PROGRAM | Mechanical Engineering

By: Michael Sebok Chair: Herbert Tanner

ABSTRACT

Heterogeneous multi-agent systems, especially those incorporating physical interaction between robotic agents, can often perform tasks that would otherwise be beyond the scope of comparable systems or require increased agent complexity.
When robots operate in hostile environments, it is desirable to reduce unit costs due to the difficulty of robot retrieval and the high risk of loss for each vehicle.
Leveraging non-trivial physical interaction between agents and agent heterogeneity, we can reduce unit cost while also increasing overall system capabilities in terms of task completion.
These features make this form of collaborative robotic system an attractive option for practical applications including search and rescue or combat scenarios.

However, the primary disadvantage of employing heterogeneous systems which incorporate physical interaction is the difficulty in generating planning and control outputs in systems with multiple agent modalities and complex dynamical models.
This is the issue that this thesis attempts to address by introducing innovations in the modeling, composition and generation of plans within a broad class of heterogeneous multi-agent systems.
The first major contribution is an adapted hybrid automata formalism to model heterogeneous systems combined with the join operation to compose systems into a team model.
This setup allows for collaborative actions to be incorporated into individual agent models which are only enabled within the team automaton through application of the join operation.
Another contribution consists of a method for abstracting these hybrid automata models onto a discrete workspace for planning while still accounting for restrictions imposed by the continuous dynamics of the system.
Finally, we introduce a process for evaluating all the possible configurations of a flexible tether within an obstructed workspace and a method for converting candidate tether geometries into discrete inputs to the corresponding system models.

The sum of these contributions is an adaptive methodology which can be utilized for modeling and planning in a broad range of hypothetical heterogeneous systems.
To demonstrate the applicability of the presented methods, we apply them to two case studies each incorporating a different form of physical interactivity between agents.
The first case study employs a UAV/UGV system with an actuated tether connecting between them.
The UAV can attach the free end of a tether to an elevated location in an obstructed environment so that the UGV may reel in the other end of the tether to move itself over obstacles.
The second case study consists of a system with two UGVs of varying modalities that can join together with permanent magnets to interact with the environment.
One UGV utilizes a Klann linkage for locomotion and the other UGV} has actuated wheels.
When joined via the magnets, the walking mechanism becomes a simple manipulator to interact with the environment and the wheeled UGV steers the joined vehicle.
These case studies illustrate two comprehensive implementations of the proposed methods, but represent only a fraction of the heterogeneous multi-agent systems that can be modeled using the processes developed for this thesis.

Back >