The contributions of this work are many and varied.
Our work has demonstrated that we can develop scalable distributed controllers for modular self-reconfiguring robots. Our controllers are distributed, based on local information only, and provably correct. Some of our algorithms are architecture-specific. Others are generic in that they can be instantiated to many different types of robot architectures. We have shown distributed algorithms of this flavor to several tasks: (1) shape morphing; (2) locomotion; (3) division of a large robot into smaller robots; (4) merging of several small robots into a large robot; (5) self-repair and (6) goal recognition.
Our work has also shows that for self-reconfiguration planning homogeneity is not that different than heterogeneity computationally (that is, the asymptotic omputation time for the plan is the same but the constants are different.) The work has also shown a close replationship between heterogeneous planning and sorting and the warehouse problem. These results lead to guiding principles for designing and building the next generation of self-reconfiguring robots.
The implementation of these algorithms on the Crystal robot has shown that it is possible to develop an autonomous robot capable of the above tasks, when the tasks are controlled in a distributed way. Our robots performed continuously for 5 days at SIGGRAPH 02 and for three additional days at AAAI 02. These demonstrations have shown that our hardware is robust and the software is efficient.
Our experiments have also pointed out some weaknesses of the Crystal hardware. The most important improvement should be in the connector design. Further improvements in the communication infrastructure and power supply would also be useful.
Our work has also demonstrated new capabilities for MEMS actuators. We have developed a robust process for post-processing MUMPS and used this process to build the first untethered scratch drive actuators. Our extensive experimentation has demonstrated that the scratch drive actuators are reliable and can move very fast.
This project led to the creation of the first self-reconfiguring robot with unit-compressible modules. We developed the hardware and the software infrastructure for controlling this robot in a distributed fashion. We developed several distributed algorithms for reconfiguration planning for this robot. We proved the correctness of, and implemented, these algorithms. We demonstrated the resulting systems at SIGGRAPH 2002 and AAAI 2002.
We started a collaboration with Prof. Duane Compton from the Dartmouth Medical School to apply the rule-based approach to local control for self-reconfiguring robots to spindle formation in cell mitosis. The process of cell mitosis is similar to the processes that lead to self-organization in several ways. In cell mitosis, a collection of rods called microtubules are organized as a spindle by the action of three different types of protein motors. This cell-level self-organization is driven by local interactions and brownian motion. We are collaborating to develop a rule-based model like those developed in our papers (ICRA 2002, DARS 2002, and IROS 2002) to this biological problem.
Self-organization is a key attribute to understanding biological systems and ultimately designing machines with higher cognition capabilities.