Periodic Reporting for period 1 - SRoCS (A Swarm Robotics Construction System)
Periodo di rendicontazione: 2019-07-01 al 2021-06-30
Swarm robotics systems have many potential applications, especially in hostile and unstructured environments where flexibility and redundancy are required to successfully perform a task. Examples of such tasks and environments could include containing a flooding river or building emergency shelter following a landslide or an earthquake. The ability to deploy such swarm robotics systems in these scenarios would likely save many human lives.
Research in swarm robotics, however, has for some time been limited to relatively simple robots whose collective task is to form and to maintain some level of spatial organisation with respect to a task and an environment. While understanding how to perform these spatial-organisational tasks is a fundamental step in this area of research, I believe that we need to push these boundaries and start investigating how swarms of robots can work together to modify their environment.
As a first step, I studied in this project how swarms of robots could cooperate to build simple structures using the BuilderBot robot and the Stigmergic Block. In particular, we wanted to study how target structures could be decomposed into sets of construction rules that can be executed by a swarm of robots.
During an experiment, these construction rules would be applied by the individual robots with respect to the current state of the environment including any partially-built structures. Construction rules such as adding building material can then be applied to an incomplete structure, advancing it towards completion.
Research in swarm robotics, however, has for some time been limited to relatively simple robots whose collective task is to form and to maintain some level of spatial organisation with respect to a task and an environment. While understanding how to perform these spatial-organisational tasks is a fundamental step in this area of research, I believe that we need to push these boundaries and start investigating how swarms of robots can work together to modify their environment.
As a first step, I studied in this project how swarms of robots could cooperate to build simple structures using the BuilderBot robot and the Stigmergic Block. In particular, we wanted to study how target structures could be decomposed into sets of construction rules that can be executed by a swarm of robots.
During an experiment, these construction rules would be applied by the individual robots with respect to the current state of the environment including any partially-built structures. Construction rules such as adding building material can then be applied to an incomplete structure, advancing it towards completion.
During the course of this project, we identified that one of the key challenges in having a swarm of robots perform construction came down to whether the individual robots had sufficient context regarding the current state of a structure to perform further construction.
The hardware used in this project was the BuilderBot robot, a mobile robot that can move around its environment, find and pick up build blocks called the Stigmergic Blocks, and place them into structures. However, the BuilderBot robot has a significant limitation in that it’s field of view has to be quite small to prevent the CPU from overheating. This limitation results in the BuilderBot not being able to see more than a few of the Stigmergic Blocks at a time, which in turn limits the complexity of the structures and reliability of the building process. Since this limit is not specific to the BuilderBot, I decided that there was value in solving this problem and investigated two ways in which it could be solved.
The first solution used the advanced capabilities of the Stigmergic Block to form a network in which adjacent blocks could exchange messages using peer-to-peer near-field communication (NFC). This network enabled the implementation of a centralized controller in one of the blocks that would then regularly communicate with the other blocks in order to monitor the state of the structure as it was being built. During construction, the centralized controller would reconfigure LEDs on the Stigmergic Blocks already in the structure to communicate with nearby robots. Upon detecting the changes in the LEDs, the robots updated their behavior. We performed experiments both using simulation and using the real hardware. Experiments from simulation have been published, while a draft of the journal article with the hardware experiments is still being prepared for submission.
The second solution involved building on a concept developed at IRIDIA known as Mergeable Nervous Systems (MNSs), in which a swarm of robots can self-assemble with individuals yielding access to their sensors and actuators to a brain robot. In contrast to previous iterations of this concept, we extended the MNS concept so that the robots no longer needed to be physically connected to each other. This control architecture would enable the BuilderBots to share and combine their fields of view and to perform construction with considerably more context than beforehand.
In order to combine the fields of view of the BuilderBots, however, the robots need to be able to maintain a common coordinate system. For this reason, we decided to target experiments consisting of both BuilderBots and custom-built quad-camera drones that could track the positions of the BuilderBots on the ground. These drones would then be able to form MNSs, enabling the fusion of the BuilderBots’ fields of view. We have published two conference papers based on results from simulation, the first focusing on the new MNSs ability to form target morphologies, the second focusing on contrasting centralized, decentralized, and MNS-based swarm systems with respect to a coverage task. An extension to the latter work has been submitted as a journal article and is currently in review. Finally, two journal articles for Science Robotics and HardwareX are currently being prepared that present real robot experiments with the MNS and the design of our open-source quad-camera drone.
Due to the situation with the global pandemic, our dissemination activities were exclusively limited to publications. This deviation from the plan was primarily due to (i) workshops and events being cancelled or not possible to do remotely and due to (ii) restricted access to our labs (which significantly delayed the testing of our new robot hardware and the running of experiments).
The hardware used in this project was the BuilderBot robot, a mobile robot that can move around its environment, find and pick up build blocks called the Stigmergic Blocks, and place them into structures. However, the BuilderBot robot has a significant limitation in that it’s field of view has to be quite small to prevent the CPU from overheating. This limitation results in the BuilderBot not being able to see more than a few of the Stigmergic Blocks at a time, which in turn limits the complexity of the structures and reliability of the building process. Since this limit is not specific to the BuilderBot, I decided that there was value in solving this problem and investigated two ways in which it could be solved.
The first solution used the advanced capabilities of the Stigmergic Block to form a network in which adjacent blocks could exchange messages using peer-to-peer near-field communication (NFC). This network enabled the implementation of a centralized controller in one of the blocks that would then regularly communicate with the other blocks in order to monitor the state of the structure as it was being built. During construction, the centralized controller would reconfigure LEDs on the Stigmergic Blocks already in the structure to communicate with nearby robots. Upon detecting the changes in the LEDs, the robots updated their behavior. We performed experiments both using simulation and using the real hardware. Experiments from simulation have been published, while a draft of the journal article with the hardware experiments is still being prepared for submission.
The second solution involved building on a concept developed at IRIDIA known as Mergeable Nervous Systems (MNSs), in which a swarm of robots can self-assemble with individuals yielding access to their sensors and actuators to a brain robot. In contrast to previous iterations of this concept, we extended the MNS concept so that the robots no longer needed to be physically connected to each other. This control architecture would enable the BuilderBots to share and combine their fields of view and to perform construction with considerably more context than beforehand.
In order to combine the fields of view of the BuilderBots, however, the robots need to be able to maintain a common coordinate system. For this reason, we decided to target experiments consisting of both BuilderBots and custom-built quad-camera drones that could track the positions of the BuilderBots on the ground. These drones would then be able to form MNSs, enabling the fusion of the BuilderBots’ fields of view. We have published two conference papers based on results from simulation, the first focusing on the new MNSs ability to form target morphologies, the second focusing on contrasting centralized, decentralized, and MNS-based swarm systems with respect to a coverage task. An extension to the latter work has been submitted as a journal article and is currently in review. Finally, two journal articles for Science Robotics and HardwareX are currently being prepared that present real robot experiments with the MNS and the design of our open-source quad-camera drone.
Due to the situation with the global pandemic, our dissemination activities were exclusively limited to publications. This deviation from the plan was primarily due to (i) workshops and events being cancelled or not possible to do remotely and due to (ii) restricted access to our labs (which significantly delayed the testing of our new robot hardware and the running of experiments).
Although this project was focused on swarm robotics construction, the majority of scientific progress made during the project was related to swarm robotics systems in general. In particular, the appreciation for centralized and decentralized control not being a dichotomy was significantly deepened through our experimental work with multi-robot systems that leverage the advantages of both control paradigms.
While the global pandemic has significantly delayed our project, the research is still ongoing and a number of impactful publications are planned for publication in the near future. These publications aim to apply the insights into how centralized and decentralized systems can be combined to swarm robotics construction.
The research in this project will contribute to how advanced robotics systems of the future will autonomously perform tasks such as building emergency shelter or performing search and rescue. These robotics systems will draw upon the research in this project and upon research from other labs which focus on aspects such as navigation of complex terrains or advanced perception and computer vision. It is the combination of all this research that will lead to advanced robotics systems that can be deployed in unstructured and hazardous environments.
While the global pandemic has significantly delayed our project, the research is still ongoing and a number of impactful publications are planned for publication in the near future. These publications aim to apply the insights into how centralized and decentralized systems can be combined to swarm robotics construction.
The research in this project will contribute to how advanced robotics systems of the future will autonomously perform tasks such as building emergency shelter or performing search and rescue. These robotics systems will draw upon the research in this project and upon research from other labs which focus on aspects such as navigation of complex terrains or advanced perception and computer vision. It is the combination of all this research that will lead to advanced robotics systems that can be deployed in unstructured and hazardous environments.