The /August issue of Offshore Engineer includes an update on patrolling robots. They are increasingly being used for inspection, maintenance and repair, and they are increasingly being operated, human-in-the-loop, in a cyber world that is a detailed digital twin of their offshore environment.
Energy Robotics, for example, has built a software platform for operating all available oil and gas ATEX certified robots, including those of ExRobotics, Taurob, and Mitsubishi Heavy Industries. It is used for planning, executing and monitoring fully autonomous operator rounds and inspection tasks with robots and drones. All data collected by the robots is integrated into one digital twin. Operators don’t have to worry about data silos being created by different robots – they all have one world view.
A recent study published by UK researchers in Energy and AI looks at current practices and future opportunities for the cyber-physical-human systems that are being developed across diverse industries, including offshore energy. Examples of potential interactions that are enabled by these systems include a ground robot might autonomously request visual data from an aerial drone, or an aerial drone could draw battery power from a ground-based robot. One robot could request a tool from another, or if it discovers a problem, it could autonomously request assistance from one with repair capabilities.
The researchers propose that the application of federated learning to this environment is worth further exploration.
IBM describes federated learning as a decentralized approach to training machine learning models. Each node across a distributed network trains a global model using its local data, with a central server aggregating node updates to improve the global model.
Federated learning eliminates the need to access or transfer large datasets. The privacy-preserving architecture of federated learning systems means that sensitive data never leaves a device. This helps minimize the risk of cyberattacks or data breaches. This is beneficial in data-sensitive environments such as energy.
Moving model learning to the edge like this means that real-time prediction is feasible. The time lag caused by sending raw data to a central server and then shipping the results back to the system is reduced, and prediction can occur even when there is no internet connection because the models are stored on the device.
What’s next? Another suggestion made by the UK researchers is to enrich digital twin environments with additional multisensory modalities such as haptics (tactile sensations) and auditory cues and feedback.
For now, robots are getting very good at understanding the environment around them. As Marc Dassler is CEO of Energy Robotics told Offshore Engineer: “You can already talk to them using ChatGPT-style large language models and say: ‘Look at all the pumps and motors and pipes. Do that three times a day and report back when something is wrong.’”
