News Photo by Kayla Wikaryasz On Thursday, Katie Skinner, assistant professor of robotics at the University of Michigan, is seen giving a lecture at the Great Lakes Maritime Heritage Center in Alpena as part of the Thunder Bay National Marine Sanctuary lecture series. Skinner talked about new technology she is helping to develop that will help shipwreck researchers capture and interpret data.

ALPENA — Katie Skinner, assistant professor of robotics at the University of Michigan, led a public lecture at the Great Lakes Maritime Heritage Center on Thursday evening, discussing new methods to map and create 3D models of shipwrecks.

According to a press release from the Thunder Bay National Marine Sanctuary, robots are becoming essential tools for underwater exploration. However, current robot systems still require human input and monitoring to complete missions.

The press release states that a challenge with current underwater exploration technologies is that camera images are degraded and acoustic sensors are difficult to interpret. Because of these challenges, robots are limited in their capacity to accurately capture the environment they are exploring.

Further, robots are unable to map complex underwater environments and structures, such as shipwrecks, in real-time.

Skinner’s presentation discussed current work that is aiming to develop methods and technology to solve some of the challenges that researchers face when trying to capture high-quality images and renderings of shipwrecks in low-quality environments with cost-effective equipment.

While using autonomous underwater vehicles (AUVs), battery powered cylindrical water exploration robots, researchers are able to capture images from the side views of the vehicle. Skinner explained that the side-view scans can create shadowing around shipwreck sites that makes it difficult to interpret the entire environment.

“In our work we want to think about how to automate this process,” Skinner said. “How can we use new tools and machine learning and artificial intelligence to interpret side-scan images (to) tell us what we have found during our surveys.”

Skinner also noted that while the military has access to the highest quality underwater technology for carrying out missions, researchers in underwater exploration often use lesser quality technology that does not have WiFi or GPS capabilities. Skinner said this makes it more difficult for researchers on the boat to communicate with the remotely operated vehicles (ROVs) and AUVs. ROVs have tethers that attach to the boat, but Skinner said that the tether can be obstructed and requires more monitoring by researchers.

Skinner explained mapping and building 3D models of shipwrecks can take up to several days after underwater exploration has been conducted. Because of the nature of their work on boats, researchers sometimes are navigating wreck sites blind.

“Sometimes we can end up with holes in our 3D models where maybe we missed collecting imagery,” Skinner said. “And this is where being able to build 3D models in real time on the boat will be really useful because we could actually see if we got a complete scan of the shipwreck before returning to shore.”

According to Skinner, part of the machine learning process involves inputting multiple examples of shipwrecks to train the AI software to accurately create real-time 3D models of shipwrecks. Part of this training included a variety of images featuring the wreck condition, sonar image quality, and terrain type. In doing so, machine learning models will be able to provide accurate output images.

However, throughout Skinner’s research, she and her team had varying results with different machine learning models.

“There’s still a long way to go here,” Skinner said. “So I think that this is a really exciting area of research for the future for us to improve on how well we can perform shipwreck segmentation from sonar data.”

Kayla Wikaryasz can be reached at 989-358-5688 or kwikaryasz@TheAlpenaNews.com.