CoCoRo (Collective Cognitive Robots), swarm of autonomous underwater vehicles, Artificial Life Laboratory, University of Graz, Graz, Austria

[Administrator]

TYOC#17/52: Lily Confinement by Bluelight

Published on Apr 26, 2015

The Year of CoCoRo 17/52: In this video we use blue-light blinks to keep the swarm together and to keep it in vicinity of the moving base station. Due to this „confinement“, the radio-controlled base-station can pull a whole swarm of Lily robots like a tail behind itself. It is important to confine the robots into specific areas in larger water bodies because the swarm requires normally a minimum connectivity among agents to work efficiently, which is achieved only with a critical minimum swarm density. Without keeping the robots in a controlled area, robots could get lost and the robot density could fall below the critical density. Thus, confinement was identified to be a critical functionality.

 

[Administrator]

TYOC#18/52: Jeff Confinement ElectricField

Published on May 3, 2015

The Year of CoCoRo 18/52: We use a submerged electrode below the CoCoRo surface station to generate a pulsing electric field underwater around this station. The Jeff robots have electrodes on their outer hull to be able to sense such fields. This way we can confine the robots into a specific area (volume) around the base station. This is important to keep the swarm together in the water, otherwise robots can get lost. We first tested this system in a pool, as it is shown in this video here.

 

[Administrator]

TYOC#19/52: Confinement Livorno

Published on May 8, 2015

The Year of CoCoRo 19/52: After having tested the electric-field confinement of Jeff robots to the base station in our pool, we went out to Livorno harbor to test it under out-of-the-lab conditions. Although our CoCoRo prototype robots were not designed to operate in salty ocean water — there is a significantly different electrical conductivity compared to freshwater — the electrical confinement worked there quite well.

 

[Administrator]

TYOC#20/52: Jeff Autonomous Docking

Published on May 17, 2015

The Year of CoCoRo 20/52: For the sake of achieving long-term energy autonomy with our CoCoRo system, we constructed a docking/undocking mechanism for Jeff robots on our surface station. First, this functionality was tested with a fixed mounted docking device, then with a docking device floating around in our pool. To say it short: Autonomous docking worked exceptionally well under all conditions.

 

[Administrator]

TYOC#21/52: Jeff Docking Livorno

Published on May 24, 2015

The Year of CoCoRo 21/52: After successful test of autonomous docking in our pool, we tested this functionality also in Livorno harbor. Despite all the waves, turbulences, wind, rain and other challenges the docking worked very reliable and fast. We could also use the docking to deploy a robot to specific sites with our radio-controlled surface station. Keep in mind: the docking and undocking is autonomous, as is also the Jeff robot as soon as it undocks from the base station.

 

[Administrator]

TYOC#22/52: Aggregation Magnets

Published on May 31, 2015

The Year of CoCoRo 22/52: In this experiment a swarm of Lily robots has to find a magnetic target on the ground. This is made difficult by water currents that make single robots drift away after they have found the target. By recruiting other robots to this location with blue-light blinks a larger group can be formed at the target site. This stabilizes the swarm at that location, thus preventing the unwanted drifting away. In addition, several weak magnetic field spots (local optima) make it even more difficult for the swarm to find and aggregate at the right (strong) magnetic target spot. Collectively the robots manage to solve the task following a very simple behavioral program and a 1-bit communication principle. This shows that very simple agents can manage to perform complex tasks together without any global knowledge available to them.

 

[Administrator]

TYOC#23/52: Lily Aggregation Experiments

Published on Jun 7, 2015

The Year of CoCoRo 23/52: Based on the algorithm we have presented last week in "The Year of CoCoRo", we conducted a series of collective choice experiments with our Lily swarm robots in two distinct environments: Open field, which asked for choosing a global magnetic optimum over several local ones, and a T-Maze setup, which was offering only a binary choice. In all experimental settings the swarm reached a clear decision in favor of the global optimum. This ability of decision making is a property of the collective, not of the individual.

 

[Administrator]

TYOC#24/52: Jeff Exploration

Published on Jun 14, 2015

The Year of CoCoRo 24/52: This video shows Jeff robots in searching a structured, highly fragmented habitat. The important issue here is the area coverage that can be achieved by the autonomous motion behavior. We found that a simple program called "correlated random walk with obstacle avoidance" does the job quite well. This program is straight forward motion altered by infrequent randomly triggered random turns and collision avoidance based on blue-light emissions and reflection at the nose of the robot. One robot covers the habitat already quickly, more robots even faster. This scalability is important to generate a useful robot swarm.

 

[Administrator]

TYOC#25/52: JEFFCAM in Pool

Published on Jun 21, 2015

The Year of CoCoRo 25/52: We put an underwater camera on one of our JEFF robots diving autonomously in a structured, in fact highly fragmented habitat. This way we want to see how it looks like to be inside of the swarm and also if we can see into the hidden places by putting a camera on an autonomous, purely "random walking" robot. Later, as we have shown already in previous weeks, we did extend this approach also to natural habitats (lakes and an ocean harbor). In the video shown here now, we tested the areal coverage of the swarm and how many interesting sites will be caught on camera.

 

[Administrator]

TYOC#26/52: Magnetic Target 4Compartments

Published on Jun 28, 2015

The Year of CoCoRo 26/52: We constructed a fragmented habitat in a pool providing 4 compartments separated by medium-height walls. One compartment contained a magnetic/metallic object acting as a search target. One JEFF robot was introduced performing a random-walk-based search pattern occasionally interrupted by a "jump-over-the-wall" behavior to allow it to switch compartment after a collision event by blue-light LEDs was detected. The autonomous robot observed its internal magnetic sensor to identify the target in case it is nearby. In this case, it stops there, sinks on top of the target and switches on its LEDs to mark it. The video shows a series of experiments to see how many successful results we can achieve in a row and how often "false positives" happen. We were quite surprised about the reliability of the behavior given the fact that the robot is only driving randomly (no specific optimized search pattern!) and no higher-cognitive capabilities (no map, no camera, no memory, no plan!) are involved.

 

[Administrator]

TYOC#27/52: Jeff target find large pool

Published on Jul 5, 2015

The Year of CoCoRo 27/52: After we tested our search algorithm for magnetic targets in the small pool (see previous movie from last week) we went out to a pool having several times this water volume. We used 4 Jeff robots this time, performing a random-walk based search program. Although we had many trouble in this long-night experiment we achieved some success in this extend search scenario. We had mostly software-problems that made the robots run too much along cyclic trajectories instead of doing an efficient random walk. This was the stuff we had to fight with in early summer 2014, later throughout the year we made significant improvements, as later videos will show. Although the target-finding rate was not perfect, the robots still found the target several times and it was the first time we operated our robots successfully at water depths below 2m.

 

[Administrator]

TYOC#28/52: All Magnetfinding

Published on Jul 12, 2015

The Year of CoCoRo 28/52: This video shows our test of "dead reckoning" performed by a Jeff robot in Livorno harbor. The robot was programmed in a way that it goes along a straight-line trajectory for some time, then performs a circular search pattern (including collision avoidance) and then, after some time, makes a direct return to the point where it was started. The robot only uses its internal magnetic compass sensor, its internal clock and blue-light LEDs for collision avoidance. No global information like GPS was involved, the robot had no absolute positioning system. We were surprised how well this very simple behavior worked despite the currents in the harbor and despite occasional collision-avoidance behaviors triggered by buoys in the harbor. The safety line (a thin fishing thread), which we applied to make sure the robot does not get lost, was actually never needed as the robot came back to us like a boomerang every time.

 

[Administrator]

TYOC#29/52: Jeff Preprogrammed Trajectory

Published on Jul 19, 2015

The Year of CoCoRo 29/52: This video shows our test of "dead reckoning" performed by a Jeff robot in Livorno harbor. The robot was programmed in a way that it goes along a straight-line trajectory for some time, then performs a circular search pattern (including collision avoidance) and then, after some time, makes a direct return to the point where it was started. The robot only uses its internal magnetic compass sensor, its internal clock and blue-light LEDs for collision avoidance. No global information like GPS was involved, the robot had no absolute positioning system. We were surprised how well this very simple behavior worked despite the currents in the harbor and despite occasional collision-avoidance behaviors triggered by buoys in the harbor. The safety line (a thin fishing thread), which we applied to make sure the robot does not get lost, was actually never needed as the robot came back to us like a boomerang every time.

 

[Administrator]

TYOC#30a/52: ANIMATION Combined Scenario No1

Published on Jul 26, 2015

The Year of CoCoRo 30a/52: In CoCoRo the search for a sunken metallic object on the ground of the waterbody was dealt with in several scenarios. The most complex one was scenario #1 which consists of the following phases: (1) The base station at the water surface moves into the habitat with docked Jeff robots and a swarm of Lily robots confined to it. (2) The Jeff robots are released, sink to the ground and search the habitat. (3) As soon as the first Jeff robot has found the target it recruits other Jeff robots to this place by blue-light LED signals, leading to a positive feedback loop aggregating more and more Jeff robots at this place. (4) In the meanwhile the Lily robots swarm out, but not as a swarm, they rather build a chain that ultimately connects the surface station to the aggregation of Jeff robots on the ground. (5) Through this "relay chain" information (blink signals or other data) can be sent from the ground swarm to the surface station (and to human operators) and vice versa. This scenario #1 combines many of the individual algorithms and functionalities that were shown in the previous videos of "The Year of CoCoRo". This video is a computer animation illustrating the phases of our "scenario #1".

 

[Administrator]

TYOC#31/52: Combined Scenario No1, Basestation Arrival

Published on Aug 2, 2015

The Year of CoCoRo 31/52: The first phase of our "combined scenario #1" is that the CoCoRo system, as a whole, enters the habitat. Then happens the only human intervention in the execution of the scenario: The swarm gets released by pressing a button performed by a human operator. This sends a WLAN signal that triggers the release of the swarm: For this very important security measure we kept the human in the loop on purpose, because a robot swarm is just a tool that a human operator, who is responsible for it, can use to achieve or facilitate a certain task. After the release of the swarm the Jeff robots sink to the ground of the water body and start their search for the metallic/magnetic target object while the Lily robots swarm out to later build the "relay chain" between the target spot and the surface station.

 

[Administrator]

TYOC#32/52: Combined Scenario No1, Targetfind

Published on Aug 9, 2015

The Year of CoCoRo 32/52: Another important phase of the "combined scenario #1" in CoCoRo is the collective search performed by Jeff robots on the ground of the water body for a sunken metallic/magnetic target object. We conducted an extra series of experiments on this phase of our "combined scenario #1" to see the efficiency and robustness of this autonomous swarming behavior. It is based on correlated random walk with collision avoidance and observation of the internal compass of the robot. After finding the object other robots are recruited to this place by horizontal blue-light signals to ultimately achieve an aggregation of Jeff robots around the target that marks this location for the Lily robots which will then form the "relay chain" to the surface station.

 

[Administrator]

TYOC#33/52: Combined 1 relay chain

Published on Aug 16, 2015

The Year of CoCoRo 33/52: In the CoCoRo system there is often a distant gap between the swarm of Jeff robots on the ground that found a search target and the base station at the water surface. Underwater communication is often difficult and limited. To bridge this gap in communication a swarm of Lily robots build a chain, which we call a "relay chain", as it serves as a relay that allow the ground swarm and the surface station to communicate with each other through light pulses, RF or electric pulses. Here we show several instances of such a relay chain in the middle phase of our "combined scenario #1" and we test information transfer in both directions by triggering either the ground swarm or the base station with a flashlight pulse. The relay chain is formed by Lily robots by executing a special "shoaling algorithm" which is inspired by fish shoaling behavior. It is based on reactions to nearest neighbors sensed passively by photodiodes on the outer hull of the Lily robots. Reception of periodic blue-light blinks allows the robots to detect each other and to align/orient relative to the neighboring configuration. Thus the relay chain formation itself is almost communication-free, as the photodiodes and blinks could be easily replaced by a video camera and object detection. It is just passive sensing and interpretation of the local environment. The only communication act performed is the relaying of the RF pulse information, which was implemented similar to slime-mould amoebas' communication strategy.

 

[Administrator]

TYOC#34/52: Relay Chain Basics

Published on Aug 23, 2015

The Year of CoCoRo 34/52: This video shows a series of experiments we performed on the building and persistency of the relay chain, which is a chain of Lily robots that connects two spatially separated points in the underwater habitat. In the first set of experiments these two end points are fixed, while in the second set of experiments they are moving, thus the relay chain formed by the robot swarm has to reconfigure and reshape. The robots forming the chain perform an algorithm inspired by the shoaling behavior of fish. They look out for blue-light blinks of neighboring robots with their photoreceptors (photodiodes). From the light intensity that falls into those sensors they infer very roughly the distance and angle of those neighbors. Based on these pieces of information they decide into which direction they turn and move. Although the mechanism involves blinking signals the algorithm can work "communication free", as an onboard camera and image processing can replace this sensing and then no signal has to be send from neighboring AUVs. Thus it can also operate on passive sensing, what is important for the scalability of the method.

 

[Administrator]

TYOC#35/52: RelayChain Communication

Published on Aug 30, 2015

The Year of CoCoRo 35/52: In this video we show a set of experiments that investigate the capability of the relay chain (formed by Lily robots) to transmit (relay) information between two spatially separated places. At one of these places we trigger a special RF (radio frequency) pulse to be emitted by a robot. Neighboring robots that receive such a pulse will also send a similar pulse, this way they relay the signal along the chain. To have control over the directionality of the signal spreading, there is also a refractory period after each relaying act in which the robot is unreceptive for the relayed signal. This system is inspired by slime mold amoebas and giant honeybees and serves very well for the underwater communication purpose.

 

[Administrator]

TYOC#36a/52: ANIMATION Relayswarm

Published on Sep 6, 2015

The Year of CoCoRo 36A/52: The "combined scenario #1", which we demonstrated in last 4 weeks on The Year of CoCoRo in detail, uses a relay chain of robots to communicate between the sea ground and the surface station. This worked well but we also investigated alternatives to this communication principle: The "relay swarm" scenario uses a swarm of Lily robots performing random walks in 3D for transmitting information about the status of the search swarm of Jeff robots on the ground. This week we show again 2 videos. This video here explains the scenario in a computer animation, the other video shows the scenario in real-world experiments.

TYOC#36b/52: CoCoRo RelaySwarm

Published on Sep 6, 2015

The Year of CoCoRo 36B/52: This video shows the real-world experiments performed in the "relay swarm" scenario. First Jeff robots search the ground of a fragmented habitat for a magnetic target. As soon as it finds the target it signals this locally with blue-light LEDs. Lily robots that also roam the habitat can pick up the signal from this Jeff robot. The info can also spread from Lily robot to Lily robot as they meet, it spreads like in an infectious process. Finally, Lily robots inform the surface station that the Jeff robot on the ground has found an interesting target. Future extensions foresee that after informing the surface station another phase starts: A second signal spreads from the surface station through the Lily robots back to the Jeff robot on the ground, ultimately make the Jeff robot to go up to the surface above the found target.