How understanding will revolutionise the way robots interact with the physical world
Robotics has been a game-changer for many industries, especially manufacturing. Robots excel in their ability to perform tasks faster, more accurately, and more safely than humans, increasing productivity and quality while simultaneously reducing costs and risks. They perform various tasks in manufacturing and, as a report by the International Federation of Robotics states, there were around 2.7 million industrial robots working in factories worldwide in 2019 (580.000 in Europe).
Nevertheless, modern robots also have many limitations. One of the biggest challenges is how to make robots perceive, understand and adapt to complex and dynamic environments. Why is this a necessity?
Most robots rely on pre-programmed, manually inputted instructions to perform their tasks. This means that they can only handle predictable and controlled situations, and they struggle with any changes or uncertainties in their surroundings. For example, if a robot needs to pick up an object from a conveyor belt, it must know the exact shape, size, position, and orientation of the object beforehand. If the object is slightly different or moved, the robot might fail to grasp it or damage it. This limits the flexibility and versatility of robots and requires constant human supervision and intervention.
Robots need more understanding to take actions that make sense.
These limitations have a big impact on the actual use of robots. For example, one of the trends in modern manufacturing is mass customisation, which is the ability to produce goods and services that meet the specific needs and preferences of individual customers. Customisation requires a high degree of flexibility and adaptability in the production process.
However, mass customisation poses significant challenges for traditional robotics, which are designed for mass production of the same standardised products. To achieve mass customisation, robots need to be able to handle a variety of tasks and products, and to adjust to changes and variations in the inputs and outputs.
Additionally, these robots should have more awareness of their surroundings, for example to prevent risky situations with humans nearby.
This requires a level of sensing, understanding, and learning that goes beyond the capabilities of conventional robots.
With CORESENSE we are focusing on research that will help robots to overcome their current limitations. CORSESE aims to engineer and deploy a deeper kind of AI; one that would allow the next generation of autonomous robotics. Research that is not based in Deep Learning, and aims to provide robots with something that current AI still is unable to do: to have understanding and common sense.
The success of the approach will be demonstrated by increased flexibility and autonomy of robots in three distinct domains, being manufacturing cells one of them. For example, the demonstrator in IMR´s Mullingar facility will see the execution of complex discrete assembly tasks by a mobile robot and a robotic assembly cell using its manipulator arm.
In this demonstrator we expect to see the following enhancements by using the CORESENSE technology:
- On-demand Operation: Enable the mobile robot to respond to different manufacturing machines on-demand, improving system uptime and reducing production bottlenecks. Ensure safe, reliable, and optimal navigation among moving obstacles.
- Augmented Inspection: Robust analysis and inspection of manufactured parts for possible defects or irregularities against their CAD models. Suggest re-iteration with exact information about the defect if discrepancies are found.
- Dynamic Manipulation Planning: Create adaptive and flexible manipulation plans to optimally deal with manufactured parts and tools of varying nature and dimensions. Requires a robust perception pipeline that can localize candidate objects within the workspace and estimate their 6D poses.
- Multi-agent Cooperation: Detect and avoid unintended collisions and modulate the robot’s behaviour to promote human augmentation for collaborative task evolution. Prioritize safety and security during collaboration over task goals and performance objectives.
