Here's how machine learning helps helicopters land on ships

But what if we told you that machine learning might soon make these challenges a thing of the past? This is the goal of a new project spearheaded by the U.S. Navy that aims to revolutionize autonomous vertical takeoff and landing (VTOL) aircraft technology.

The US Navy is trying to find a new way to land helicopters on ships without human intervention, and it has roped in researchers from Texas A&M University to develop a solution.

In a press release, it said, the researchers are working on a novel approach that combines an optimal aircraft design with a trained machine-learning algorithm to enable the fully autonomous landing of an aircraft.

Breaking the code of Horizon Bar landings

Dr. Moble Benedict, associate professor in the Department of Aerospace Engineering at Texas A&M, explained a fascinating point about helicopter pilots. “They don't focus on the moving deck when landing. Instead, they're trained to keep their eyes on a specialized piece of equipment known as the 'horizon bar'," he said. This gyro-stabilized, lighted strip provides an artificial horizon, acting as a point of reference for the pilot.

While previous studies directed efforts toward tracking the ship’s ever-moving deck using cameras, GPS, and lidar as mentioned earlier, Benedict, and his co-PI Dr. Dileep Kalathil, have taken a fresh route. They aim to automate the entire landing process by replicating a pilot's behavior while focusing on the horizon bar.

A reinforcement learning process

What makes their approach truly innovative? "Reinforcement learning," answered Dr. Kalathil, assistant professor in the Department of Electrical and Computer Engineering. This class of machine learning is the building block for the control algorithm directing the autonomous systems. "Our algorithm will be so precise that even when the vehicle changes its course or faces heavy winds, it will track the horizon bar with pinpoint accuracy," Kalathil added.

This fusion of aerospace engineering and electrical and computer engineering isn't just theoretical. Benedict and Kalathil have already demonstrated successful tracking and safe landings of unmanned aerial systems (UAS) in challenging conditions—from windswept terrains to fog-blanketed vistas.

Customized VTOL aircraft designs

Benedict is channeling his rotorcraft expertise into crafting VTOL aircraft that are gust-tolerant and efficient, potentially featuring foldable wings for a seamless transition from vertical takeoffs to fixed-wing cruising. These designs will be tested through simulations, wind-tunnel analyses, and real-world flight tests.

Kalathil, on the other hand, is focusing on robust algorithmic solutions. "The real challenge lies in the unpredictable behavior of rough seas," he says, revealing plans to use wind sensors for real-time adjustments. "It allows us to counteract specific conditions, bridging the simulation-to-reality gap that has stymied previous efforts."

A triple-play for the U.S. Navy

The Navy is essentially on the lookout for a triple threat—an aircraft that's runway-independent, fuel-efficient for extended flights, and capable of safe, automated landings on a moving ship. This three-year project, funded by the U.S. Navy, could very well be the answer to that quest.

Benedict and Kalathil are not just working in their isolated domains but are also collaborating across disciplines. Kalathil is even contemplating a collaborative console that could control multiple UASs, heralding a new era of swarm intelligence in naval operations.

As they forge ahead, one thing is clear: This interdisciplinary endeavor stands to redefine not just naval operations but also how we think of machine learning's role in tackling real-world challenges.