One of the primary areas of focus for the Space Robotics Group, as well as many other robotics labs, is swarm or multiagent technology. This essentially is the cooperation of two or more robots functioning together to complete a task. Having more than one robot work toward a goal can be beneficial in a number of ways, namely, more working together can complete the task faster and at times a single robot may not be able to complete something on its own, and multiple robots are necessary.
For the better part of the last decade, the Space Robotics Group has been involved in research toward extraterrestrial mining. This is an area of interest because of plans to build manned bases on the Moon or Mars. Having direct human involvement in physical labour in the harsh environments of other planetary bodies requires significant resources. By having robots tackle these tasks will reduce the danger and resources used in the long run pertaining to developing extraterrestrial bases.
The Space Robotics Group has had several rover teams dedicated to research in this field. Initially, a fleet of ten Argo-Class rovers were used for two dimensional resource allocation simulations and experiments. These small rovers were equipped with active vision systems and various other sensors to interact with their environment as well as each other.
More recently, several members of the Space Robotics Group ventured to the United States to demonstrate the capabilities of the newer Musketeer rovers. This larger platform was used to perform full scale three dimensional resource allocation and excavation. More on this project can be found in the Recent Projects section.
Rather than being confined to the ground, the Space Robotics Group is also taking flight. Recent developments in materials and in microfabrication techniques have allowed the development of smaller and smaller aircraft. In the early days of this field, researchers worked with mainly fixed wing or rotor craft configurations. However, they quickly found that these aircraft were extremely inefficient because of the low Reynolds number flows dictated by the small scale of the aircraft. Taking a step back, researchers turned to nature for inspiration and began to develop aircraft based on its best fliers: birds and insects.
Continuing in this vein, the Space Robotics Group is developing a dragonfly-inspired MAV. The primarily carbon fibre prototype currently has 1 degree of freedom actuated by a piezoelectric bimorph and weighs less than one-tenth of a penny. When completed, the MAV is projected to be 40 mm long, have a wing span of 60 mm and a mass of 140 mg.
We envision a future where a swarm of these small MAVs can be used for search and rescue by flying over rubble where ground-based rovers have difficulty traversing. A swarm of dozens of these robotic dragonflies would be more effective in covering a large area than a single traditional, more expensive UAV. A number of applications such as surveillance or exploration could be ideal tasks for these small and nimble robots.
It is not only nature's forms that inspire us as roboticists, but also its control strategies and developmental mechanisms. Particularly with regards to collective multiagent systems, designing controllers by hand using knowledge-based approaches has proven difficult. We find that the process by which local behaviours combine into global coordination is counterintuitive. And yet we need only look to the impressive excavation and gathering achievements of an ant colony to see that the problem is soluble: indeed, evolution has solved it.
The Space Robotics Group is involved in several projects to harness evolution's propensity for generating complexity from many simple elements...
Perhaps the most complicated system that nature has engineered is the brain, with much neuroscientific evidence suggesting that this system exhibits chaotic dynamics. Current work seeks to model brain chaos and to characterize episodic memory storage and retrieval phenomena. This may allow future robots to operate flexibly in novel situations based on past experiences.
Furthermore, evolutionary algorithms are being used to evolve systems of ordinary differential equations that act as robotic controllers. Such controllers are implicitly designed to demonstrate development, learning and strong generalization capabilities, including the capacity to self-calibrate in situ.
The dynamics of the evolutionary process are being charted using concepts from information theory. By investigating the build-up of mutual information between simulated evolving agents and their environment, it may be possible to extract bulk metrics that describe the direction and qualities of evolution itself.
The video below demonstrates the capabilities of an 'artificial neural tissue' developed by the Space Robotics Group and trained by an evolutionary algorithm. A multiagent system of rovers is tasked with autonmously gathering resources, initially scattered about the blue-bordered workspace, into the orange-bordered collection area.