The industry and public sector are pushing for robotic solutions with high level of autonomy that can operate safely and efficiently in increasingly complex operations. And great strives have been done to realize such systems. In particular, research and development on stand-alone automated functions for robot acting (e.g., grasping, collision avoidance, path planning), as well high-level automated planning of missions have come a far way. However, even for simple missions, the planning complexity quickly explode. Moreover, high-level planning is often performed as an "offline" process where the environment a robot operates in is assumed to be static, i.e., "standing still". However, the environment robots and we humans operate in is in many cases not static.
To meet these demands, we will in the ROBPLAN project develop new methods within a scientific area in AI called automated planning and acting (AI planning). AI planning is about enabling robots to coordinate their capabilities and handle changes and constraints in their environment to successfully achieve mission goals. To achieve such AI planning we will build on techniques from symbolic AI approaches (i.e., classical AI), which involve structured prior knowledge about the world. And we will enhance these techniques by non-symbolic AI (e.g., machine learning), which can be used to learn from new data and adjust to changes in the world.
We will demonstrate project results on industry-relevant use cases with mobile robots and unmanned aerial vehicles (UAVs) for inspection and/or emergency handling. The project goals will be achieved through close collaboration between a research institute (SINTEF, project lead), a university (NTNU), an industry end-user (Equinor) and an industrial supplier (Scout Drone Inspection). Project results will create opportunities for robotics within a range of sectors dealing with inspection and maintenance, as well as other areas such as within agri-food and healthcare.
In ROBPLAN we will move the research front by developing methods for AI-based planning and acting - AI planning - to enable robust autonomous robot missions. We will demonstrate results on industrial use cases with mobile manipulators and Unmanned Aerial Vehicles (UAVs) for inspection and maintenance (I&M). The global I&M market is estimated at 450 billion EUR.
The industry and public sector are pushing for robotic solutions with high level of autonomy that can operate safely and efficiently in increasingly complex operations. And great strives have been done to realize stand-alone automated functions for robot acting (e.g., grasping, collision avoidance), as well as for high-level automated planning of missions. However, even for simple missions, the planning complexity quickly explodes, and high-level planning is often performed as an "offline" process where the world is assumed to be static. To enable real-life fully autonomous single- and multi-robot missions, we need robots that can balance long-term planning with the ability to react to immediate events. To meet this need we will develop methods in the ROBPLAN project to tightly combine planning and acting by building on techniques from symbolic AI approaches enhanced by non-symbolic AI. Moreover, we will develop methods for distributed robot decision-making during planning and acting to enable multi-robot autonomous missions with and without humans in the loop.
We will demonstrate results on I&M use cases within the oil and gas industry, but project results will also be applicable in other application domains where autonomous robots are increasingly deployed and can benefit from a higher level of autonomy in operations; agri-food, healthcare and manufacturing. The combination of new scientific results beyond the state of the art and real-life demonstration will increase impact of results both in research and industry.