Today's oceans contain 26-66 million tons of waste, with approximately 94% located on the seafloor. The SeaClear project aims at automating the process of searching, identifying, and collecting marine litter, using a team of autonomous robots that work collaboratively.
To find more about each step of the procedure, click on the headers below
Our unmanned surface vehicle, the SeaCat from SubseaTech, scans the area of interest of the sea bottom using a multibeam echosounder, which produces a 3D bathymetry map of the bottom. This serves as a reference map to which all other information about litter will be added. Some large litter like tires or pipes may already be detectable from the bathymetry data, in which case this litter is already marked on the map.
The SeaCat also serves as the "mothership" of the system: all other robots deploy from and return to it. Robots communicate to the USV and get power from it via tethers. The computational resources required for sensing, control, and artificial-intelligence components are also hosted by the SeaCat.
When the water is sufficiently transparent, an unmanned aerial vehicle (UAV or drone) searches for litter from the air. Larger litter pockets are expected to be identifiable in this way, and inform the more detailed search using the underwater robot in the next step. We are also investigating whether we can find correlations between surface and underwater litter pockets.
Our drone is a DJI Matrice M210 RTK, which we are modifying by adding a power and data tether to connect it with the surface vehicle.
In murky waters, the UAV still remains useful by scanning the surrounding area for obstacles.
A small unmanned underwater vehicle (UUV) is deployed from the USV and performs close-up targeted scans of the sea bottom to find smaller litter. To this end, it uses a camera and a forward-looking sonar, together with possibly other sensors such as metal detectors. Identified litter is placed on the reference map.
Litter is identitied with artificial intelligence, deep-learning object recognition techniques. These deep networks are trained so as to differentiate litter from sea-life and thereby ensure the system only collects what it should.
Our observation UUV is the miniTortuga from SubseaTech.
A larger brother of the observation UUV, called the Tortuga, goes to each piece of litter on the map and grabs it with a gripper that is custom-made to interface with the Tortuga. This gripper is equipped with a suction device that will help with picking up litter in difficult circumstances, such as when it is lying among plants. Each piece of litter is reacquired with high accuracy and then picked up.
To plan the paths and control the motion of both the observation and collection UUVs, we exploit intelligent techniques such as reinforcement learning and data-driven control.
A basket is deployed from the USV, and the collection UUV Tortuga takes each piece of litter to deposit it in the basket for transportation to the shore. The basket opening is specially designed to interface effectively with the gripper, and to prevent floating litter to escape back into the water.
The basket is not just a passive component, but actively sends signals to help the collection UUV localize itself relative to the opening.
The SEACLEAR kick-off meeting took place on the premises of the project leader, TU Delft. During this meeting, we refined the objectives and tasks of the project with all the partners.
As with any complex system, we must define the use cases before developing any of the technology involved. For the duration of this project, we will focus on two use cases, one being the tourism sector and the other port management.
At the core of our system are the robots that will do the actual work. SEACLEAR required the development of a new autonomous surface vehicle (SeaCAT), and a new underwater observation robot (MiniTortuga). Furthermore, a reliable way of deploying and retrieving all the robots is necessary.
Sensing is the way our robots locate themselves and find litter. In SEACLEAR, we are mainly relying on cameras and sonars to search for litter, and we have to integrate these sensors with the rest of the equipment.
A key component for the collection of litter is the system's gripper, which needs to be mounted under the collection ROV. The gripper should be suitable for collecting a wide range of litter found on the ocean floor, while ensuring that no marine life forms are harmed.
The collection basket is the place where the litter will be stored until it is safely transported to the shore. It must be compatible with the gripping system and must ensure that no litter escapes once collected.
Due to signal loss underwater, the ROVs cannot rely on satellite positioning systems for their localisation. Therefore, we need to develop solutions for fusing different sensor data to estimate the position of the robots underwater. Furthermore, we develop advanced techniques for controlling the trajectories of the robots efficiently while in water.
To collect litter, it must first be detected. To do this, we rely on machine learning techniques with deep neural networks. A key objective of these algorithms is to correctly distinguish between marine litter and marine life forms, so that we only collect what we intend to.
To coordinate the operation of the robots, we need sufficiently accurate knowledge of the ocean floor bathymetry. Identified litter must moreover be placed on this map, so that the robots know where to go to collect it. Our goal is to build this combined map of the ocean floor in an energy and time efficient way.
Besides the trajectory control needed for the robots, the SEACLEAR system must define higher level tasks for them. This includes the paths they will follow to detect the litter more efficiently, but also the order that the detected litter will be collected afterwards.
SEACLEAR is aiming to combine several individual components together, with a considerable amount of data that need to be shared between each component of the system. Therefore, we aim to develop reliable and efficient communication systems so that the computational intensive tasks can take place where it is more suitable.
The end goal of this project is to demonstrate its capabilities on different sites, with different needs. For SEACLEAR we have identified two locations for performing these demonstrations: The Dubrovnik area (tourist use case), and the Hamburg Port (port use case).
Our goal is to make the SEACLEAR system commercially viable. We aim at exploiting the results of this project both by selling the system and by providing it as a service to local authorities, ports, or other stakeholders.
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 871295.
The SEACLEAR project spans four years, running from January 1st, 2020 to December 31st, 2023.