This paper presents a Multiagent Systems based design approach for designing a self-replicating robotic manufacturing factory in space. Self-replicating systems are complex and require the coordination of many tasks which are difficult to control. This paper presents an innovative concept using Multiagent Systems to design a robotic factory for space exploration. Specifically presented is an approach for coordinating a conceptual model of a self-replicating system. The arrival of a set of agents on an unknown planet is simulated, whereby these simple agents will expand into a self-replicating factory using the regolith gathered from the surface of the planet. NASA is currently investing in space exploration missions that consider using the resources on the surface of other planets, asteroids or satellites. The challenge of the project is in the implementation of a learning algorithm that allows a large number of different agents to complete simultaneous tasks in order to maximize productivity. The simulation in this work is able to present the coordination of the agents during the construction of the factory as the parameters of the learning algorithm are changed. System performance is measured with a pre-programmed method, using local and difference rewards. The results show the advantage of using a learning algorithm to better build the robotic factory.

Turning a robot, particularly an under-actuated bipedal humanoid robot, is challenging. Several methods proposed in the literature for producing human-like motion in such robots are innovative but are limited in their range of motion. This paper presents an approach to control the orientation of a robot using a control moment gyroscope (CMG). A demonstration platform is developed to test this concept and physical experiments are conducted to determine the prototype's turning range and performance. This concept is then extended to a backpack mount where trials are conducted using human subjects to estimate the performance of the system that can potentially be used to turn bipedal humanoid robots.

With the increase in computer-controlled hybrid machining (e.g. mill-turn machines), one needs to discern what features of a part are created during turning (i.e. with a lathe cutter) versus those created by milling. Given a generic part shape, it is desirable to extract the turnable and non-turnable features in order to obtain feasible machining plans. A novel approach for automating this division and for defining the resulting turning operations in a hybrid process is proposed in this paper. The algorithm is based on identifying the dominant rotational-axis and performing several non-uniform lateral cross-sections to quickly generate the “as lathed” model. The part is then subtracted from the original model to isolate the non-turnable features. Next, resulting model and features are translated to a label rich graph and fed into a grammar reasoning tool to produce feasible manufacturing plans. The setup design is also studied against the tolerances specified by the designer. Performance of the algorithm has been tested on several examples ranging from simple to complex parts.

Assembly planning is an important task for manufacturing industrial products. The ability to automate and optimize the process is crucial as it can improve the production time, cost and efficiency. In general, the planning procedure involves two main reasoning stages, one is based on geometric reasoning and the other is based on AI tree-search. In this article, a novel approach to automate the geometric reasoning stage for large assembly systems is proposed. This technique provides a fast and accurate measure of capturing feasible spatial interactions and relationships between elements of a CAD assembly with no manual interventions. These results are then used by a suite of AI planning tools to generate optimal assembly sequences. The algorithm has been tested on a variety of examples resembling real products and results show the efficiency and effectiveness of the algorithm.