EBN- Robot capable of swimming underwater have seen great progress in recent years. However, underwater handling remains one of the most challenging tasks due to complex fluid dynamics and unpredictable conditions.
To overcome these challenges, a team of researchers at Columbia University is developing AquaBot, an underwater robot capable of performing advanced handling tasks completely autonomously.
Underwater handling
Water generates unpredictable forces that impede precise movements, making it even more challenging for robots to manipulate objects underwater. Traditionally, underwater robotic systems have relied on human operators to guide their operations, limiting their efficiency and scalability.
To overcome this, Columbia University researchers have designed the AquaBot, which leverages artificial intelligence and self-learning techniques to autonomously perform a variety of underwater tasks.
Aqua Boat Design
The AquaBot is built on an underwater drone called the QYSEA V-EVO with the addition of a parallel handle and two cameras to enable the robot to collect underwater images and perform handling tasks. The team developed advanced software that guides the AquaBot’s operations, allowing it to learn vision-motor policies that link visual inputs to motor commands.
The AquaBot training involved two main phases. In the first phase, the researchers recorded human operators performing various underwater tasks, such as grasping and sorting objects. They then used these recordings to train the robot’s vision-motor policy, which mimics human adaptation. Reducing the decision horizon improved the robot’s response speed, allowing it to adapt to unexpected underwater conditions.
In the second phase, the team introduced “self-improvement” technology, allowing AquaBot to improve its skills using feedback from its own performance to accelerate learning and improve efficiency.
Real-world applications and achievements
To test the AquaBot’s capabilities, the researchers ran a series of real-world tests that included tasks like grasping rocks, sorting trash, and retrieving large objects that mimicked human bodies. The AquaBot proved adept at these tasks, completing them 41 percent faster than human operators.
Tests also showed that AquaBot could generalize its skills to new tasks and unfamiliar environments. For example, the robot successfully grasped previously unseen objects, sorted trash into appropriate containers, and retrieved large objects in rescue scenarios.
