by Nathaniel Whelan
With the invention of the Canadarm, Canada has been at the forefront of space robotics since the early 1980s. This iconic 50 foot arm supported US space shuttle missions for 30 years and was built to deploy, manoeuvre, and capture payloads.
Almost a decade after its final flight, recent Carleton PhD grad Collins Ogundipe jumped on this Canadian legacy with the goal of pushing the field even further. Funded by the European Space Agency, he engaged in cutting-edge machine learning research designed to enhance the capabilities and efficiency of space robotic manipulators.
Collins’ Journey to Carleton University
Originally from Nigeria, Collins first graduated from the National Aerospace University in Ukraine, after which he received a full scholarship to study Aerospace Science and Engineering at the University of Toronto.
Following some time in the industry, Collins decided to pursue his PhD. With the ongoing evolution toward machine learning, he actively sought out labs that would allow him to engage in artificial intelligence within the context of space robotics.
His search took him to Dr. Alex Ellery with the Department of Mechanical and Aerospace Engineering at Carleton. Dr. Ellery is a Canada Research Professor whose research interests include applied space robotics and space-based manipulators, among other things.
Speaking about this partnership, Collins said:
“I was looking for something niche and multidisciplinary. I found a couple of options, but for me, there was only one fit. Dr. Ellery offered me the best research project opportunity.”
During their initial conversations, Dr. Ellery shared some ideas with Collins, igniting what would become a fruitful 3 year collaboration.
Advancing Space Robotics
Space robots, such as manipulator arms, are teleoperated from Earth because of limited computational resources on board the spacecraft. However, the inherent time delay in the signal transmission impedes the application of these manipulators. This makes them less reactive, often resulting in an error between the command and the execution of the task.
“The goal might be very tiny, if you want to screw a bolt for example, but it’s still very important,” Collins explained. “When you want to move the manipulator from point A to B, even if you calculate the trajectory, your actual position in space might not be the desired position.”
To start, Collins looked into neural networks, a model of artificial intelligence that teaches computers to process data in a way that is inspired by the human brain.
Imagine an Amazon package. You think it is light, but when you lift, it is actually quite heavy. Nevertheless, your muscles adjust quickly, allowing you to pick it up. The human cerebellum is responding to this sensory change; it is a process that happens almost instantaneously.
“As an engineer, I looked at how the motor cortex in the brain works. With a neural network, our aim was to develop a robot that mimics that architecture and provides compensation when they encounter errors.”
In other words, it would make that “muscle” adjustment itself.
This prompted Collins to adopt a predictive neural network. Because operators do not have the luxury to train controllers on board the spacecraft, the idea is to deploy models that have been pretrained on Earth to circumvent issues and adapt to unseen scenarios. The robot would update itself based on how it responds to the environment, giving way to a system that is ultimately more compliant.
Funding and the Future
To support his project, Collins secured funding from the European Space Agency (ESA). With a current annual budget of €7.08 billion, the ESA is a 22-member intergovernmental body dedicated to advancing Europe’s space capabilities and ensuring that investment in space continues to yield benefits to the citizens of the world.
Collins and Dr. Ellery worked on the proposal together, with the latter serving as principal investigator. The financial backing from the ESA provided Collins with the necessary resources to build a model and run countless simulations to validate different scenarios and showcase how pretrained space robotics can be adopted.
“Having the ESA buy into the idea of my research and supporting it in this way has given me room to do more than what others might have been able to accommodate.”
In the early days of the project, the ESA offered for Collins and Dr. Ellery to come down and use their facilities, but Covid-19 interrupted those plans. Even so, Collins acknowledged that the door is still open, offering a path forward to collaborate and conduct further experiments based on his initial research.
Conclusion
Collins’ project is tailored to space missions, but could have terrestrial applications as well in any field where compliant grasping is required, such as manufacturing.
In addition to research, Collins has a passion for teaching, winning the 2020/21 Outstanding TA award after being nominated by his students and peers.
To learn more about the research activities happening in the Department of Mechanical and Aerospace Engineering, please visit their website.