Melvin Gauci

Self-organized aggregation without computation

This paper presents a solution to the problem of self-organized aggregation of embodied robots that requires no arithmetic computation. The robots have no memory and are equipped with one binary sensor, which informs them whether or not there is another robot in their line of sight. It is proven that the sensor needs to have a sufficiently long range; otherwise aggregation cannot be guaranteed, irrespective of the controller used. The optimal controller is found by performing a grid search over the space of all possible controllers. With this controller, robots rotate on the spot when they perceive another robot, and move backwards along a circular trajectory otherwise. This controller is proven to always aggregate two simultaneously moving robots in finite time, an upper bound for which is provided. Simulations show that the controller also aggregates at least 1000 robots into a single cluster consistently. Moreover, in 30 experiments with 40 physical e-puck robots, 98.6% of the robots aggregated into one cluster. The results obtained have profound implications for the implementation of multi-robot systems at scales where conventional approaches to sensing and information processing are no longer applicable.

Occlusion-Based Cooperative Transport with a Swarm of Miniature Mobile Robots

This paper proposes a strategy for transporting a large object to a goal using a large number of mobile robots that are significantly smaller than the object. The robots only push the object at positions where the direct line of sight to the goal is occluded by the object. This strategy is fully decentralized and requires neither explicit communication nor specific manipulation mechanisms. We prove that it can transport any convex object in a planar environment. We implement this strategy on the e-puck robotic platform and present systematic experiments with a group of 20 e-pucks transporting three objects of different shapes. The objects were successfully transported to the goal in 43 out of 45 trials. When using a mobile goal, teleoperated by a human, the object could be navigated through an environment with obstacles. We also tested the strategy in a 3-D environment using physics-based computer simulation. Due to its simplicity, the transport strategy is particularly suited for implementation on microscale robotic systems.

Why `GSA: a gravitational search algorithm' is not genuinely based on the law of gravity

This letter highlights a fundamental inconsistency in the formulation of the Gravitational search algorithm (GSA) (Rashedi et al., Inf Sci 2232–48, 2009). GSA is said to be based on the law of gravity, that is, candidate solutions attract each other in the search space based on their relative distances and ‘masses’ (qualities). We show that, contrary to what is claimed, GSA does not take the distances between solutions into account, and therefore cannot be considered to be based on the law of gravity.

A strategy for transporting tall objects with a swarm of miniature mobile robots

This paper proposes a strategy for transporting a tall, and potentially heavy, object to a goal using a large number of miniature mobile robots. The robots move the object by pushing it. The direction in which the object moves is controlled by the way in which the robots distribute themselves around its perimeter - if the robots dynamically reallocate themselves around the section of the objects perimeter that occludes their view of the goal, the object will eventually be transported to the goal. This strategy is fully distributed, and makes no use of communication between the robots. A controller based on this strategy was implemented on a swarm of 12 physical e-puck robots, and a systematic experiment with 30 randomized trials was performed. The object was successfully transported to the goal in all the trials. On average, the path traced by the object was about 8.4% longer than the shortest possible path.

Evolving Aggregation Behaviors in Multi-Robot Systems with Binary Sensors

This paper investigates a non-traditional sensing trade-off in swarm robotics: one in which each robot has a relatively long sensing range, but processes a minimal amount of information. Aggregation is used as a case study, where randomly-placed robots are required to meet at a common location without using environmental cues. The binary sensor used only lets a robot know whether or not there is another robot in its direct line of sight. Simulation results with both a memoryless controller (reactive) and a controller with memory (recurrent) prove that this sensor is enough to achieve error-free aggregation, as long as a sufficient sensing range is provided. The recurrent controller gave better results in simulation, and a post-evaluation with it shows that it is able to aggregate at least 1000 robots into a single cluster consistently. Simulation results also show that, with the recurrent controller, false negative noise on the sensor can speed up the aggregation process. The system has been implemented on 20 physical e-puck robots, and systematic experiments have been performed with both controllers: on average, 86-89% of the robots aggregated into a single cluster within 10 minutes.

A coevolutionary approach to learn animal behavior through controlled interaction

This paper proposes a method that allows a machine to infer the behavior of an animal in a fully automatic way. In principle, the machine does not need any prior information about the behavior. It is able to modify the environmental conditions and observe the animal; therefore it can learn about the animal through controlled interaction. Using a competitive coevolutionary approach, the machine concurrently evolves animats, that is, models to approximate the animal, as well as classifiers to discriminate between animal and animat. We present a proof-of-concept study conducted in computer simulation that shows the feasibility of the approach. Moreover, we show that the machine learns significantly better through interaction with the animal than through passive observation. We discuss the merits and limitations of the approach and outline potential future directions.

Shepherding with robots that do not compute.

We examine the problem solving capabilities of swarms of computation- and memory-free agents. Each agent has a single line-of-sight sensor providing two bits of information. The agent maps this information directly onto constant motor commands. In previous work, we showed that such simplistic agents can solve tasks requiring them to organize spatially (multi-robot aggregation and circle formation) and manipulate passive objects (clustering). In the present work, we address the shepherding problem, where the computation- and memory-free agents—the shepherds—are tasked to gather and move a group of dynamic agents—the sheep—towards a pre-defined goal. The shepherds and sheep are modelled as e-puck robots using computer simulations. Our findings show that the shepherding problem does not fundamentally require arithmetic computation or memory to be solved. The obtained controller solution is robust with respect to sensory noise, and copes well with changes in the number of sheep.

Programmable Self-disassembly for Shape Formation in Large-Scale Robot Collectives

We present a method for a large-scale robot collective to autonomously form a wide range of user-specified shapes. In contrast to most existing work, our method uses a subtractive approach rather than an additive one, and is the first such method to be demonstrated on robots that operate in continuous space. An initial dense, stationary configuration of robots distributively forms a coordinate system, and each robot decides if it is part of the desired shape. Non-shape robots then remove themselves from the configuration using a single external light source as a motion guide. The subtractive approach allows for a higher degree of motion parallelism than additive approaches; it is also tolerant of much lower-precision motion. Experiments with 725 Kilobot robots allow us to compare our method against an additive one that was previously evaluated on the same platform. The subtractive method leads to higher reliability and an order-of-magnitude improvement in shape formation speed.

Recent Advances in Swarm Robotics

We overview recent advances in swarm robotics, particularly concerning the control of groups of autonomous robots and the identification of swarming rules through observation. Swarm intelligence is the study of systems of spatially distributed individuals that coordinate their actions in a self-organised manner and thereby exhibit complex collective behaviour [1]. Figure 1 Examples of tasks demonstrated by swarms of autonomous mobile robots. Left: A group of robots transporting a comparatively large object towards a goal location (red cylinder). Centre: A group of robots having organised themselves into a single cluster. Right: A group of robots clustering objects, simulating a litter collection scenario. In the first part of the talk, we present our recent advances in controlling groups of robots. The platform being used is the EPFL miniature mobile robot e-puck. It is shown that some tasks can be solved by swarms of robots with severely limited abilities. For example, in order for a group of robots to transport a tall object (see Figure 1, Left), the robots do not fundamentally require to communicate with each other in an explicit way [2]. Rather it is sufficient if they can discriminate between the object, the goal and the remainder of the environment. In order for a group of robots to gather in a single place [3] (see Figure 1, Centre), or cluster objects that are initially dispersed [4] (see Figure 1, Right), it was found that the robots do not fundamentally require arithmetic computation. Such tasks can be solved by robots that use a binary sensor to trigger one of two possible actions, without the need to store information during run-time. In the second part of the talk, we present a method that is able to identify models (parameters) of individuals, for example, when part of a swarm, through observation and interaction [5,6]. This method does not require any pre-defined metric to gauge the resemblance of models to observed individuals. Research challenges in swarm robotics are manifold. They include: · Using swarm robotic systems in real-world scenarios (e.g. Precision farming, waste management); · Formalising the design, implementation and verification process; · Miniaturising swarm robotic systems, for example, for applications in healthcare; · Improving tools to learn about and influence natural swarms. References 1. C Blum and R Groß. Swarm intelligence in optimization and robotics: A concise introduction, Springer Handbook of Computational Intelligence, Springer (in press) 2. J Chen, M Gauci, R Groß. A strategy for transporting tall objects with a swarm of miniature mobile robots, ICRA 2013, 863-869 http://dx.doi.org/10.1109/ICRA.2013.6630674 3. M Gauci, J Chen, W Li, TJ Dodd, R Groß. Self-organized aggregation without computation, International Journal of Robotics Research, OnlineFirst http://dx.doi.org/10.1177/0278364914525244 4. M Gauci, J Chen, W Li, TJ Dodd, R Groß. Clustering objects with robots that do not compute, AAMAS 2014, 421-428 http://dl.acm.org/citation.cfm?id=2615800&CFID=454901457&CFTOKEN=66407710 5. W Li, M Gauci, R Groß. A coevolutionary approach to learn animal behavior through controlled interaction, GECCO 2013, 223-230 http://dx.doi.org/10.1145/2463372.2465801 6. W Li, M Gauci, R Groß. Coevolutionary learning of swarm behaviors without metrics, GECCO 2014 (in press)